AI Journey from India to London to San Francisco

"I Moved from India to London to San Francisco to Build THIS"! (2 Years of AI Learnings)

Estimated read time: 1:20

Summary

This captivating video discusses the journey of two young AI innovators who moved from different parts of the world to San Francisco to build a powerful AI assistant. The creators, aged 19 and 24, built an AI assistant called AMX, which integrates with daily tools and enhances productivity. They share their experiences of transitioning across geographies, the challenges faced in the AI landscape, their approach to building this innovative product, and their philosophy on open-sourcing technology. This video is a rich resource for anyone looking to understand the intricacies of building AI solutions in today's world. The insights shared reflect their commitment to solving real-world problems with technology, making this a compelling watch for aspiring tech entrepreneurs.

Highlights

Two young founders embark on a global journey to create groundbreaking AI technology! 🌐
AMX, the AI assistant, was built in just three months and offers unprecedented developmental speed! ⚡
Key insights into tech development and entrepreneurship in one engaging conversation! 🎤
Exploration of the balance between cutting-edge technology and user-centric product design. 🤖
Epic tales of innovation, from coding journey beginnings to real-world impact! 🛠️

Key Takeaways

AI and innovation know no age limits! Young entrepreneurs are making waves! 🌊
Adapting to new environments can be challenging, but it’s a key to growth and innovation. 🌍
Open sourcing can drive transparency and improve collective innovation in tech! 💻
Building an AI company is tougher than it seems, even for those who've succeeded before! 😅
AMX stands out by integrating seamlessly with existing tools to assist users like a silent power-up! 🚀

Overview

In this illuminating conversation, we follow the remarkable journeys of two young innovators, Sansar and Aron, who moved from India and Kazakhstan respectively, through London, and finally to San Francisco. Their mission: to build AMX, a personal AI assistant designed to seamlessly integrate into everyday digital tools. Despite their youthful ages, these entrepreneurs bring valuable lessons from their cross-continental experiences to the forefront.

Sansar and Aron discuss the development of AMX, which was accomplished in a rapid three-month timeframe. The AI assistant is designed to work invisibly across various platforms, streamlining users’ digital and organizational workload. Their technological prowess is evident in the creation of a solution that speeds up performance by substantial margins compared to existing technologies like Langchain.

Moreover, the conversation delves into their philosophical approach to open-sourcing their project. By choosing transparency and collective input over proprietary restrictions, Sansar and Aron aim to foster an inclusive and innovative tech community. Their story is not just about building a product but reshaping how technology intertwines with personal productivity and innovation.

Chapters

00:00 - 00:30: Introduction This chapter discusses the extent to which AI-generated code is used in the codebase, estimating it to be about 80 to 90%. It touches upon the challenges faced when starting an AI company, acknowledging that it was more difficult than initially anticipated. There is also a discussion on the philosophy and considerations behind open sourcing an AI company, especially in the context where many AI agents remain proprietary.
00:30 - 01:00: Overview of Jarvis AI Assistant The chapter provides an overview of Jarvis, a personal AI assistant developed by a 19-year-old and a 24-year-old founder. It details the founders' journey from India and Kazakhstan to the UK and San Francisco, highlighting their ability to get funding despite starting without a concrete idea, leveraging only their experience.
01:00 - 01:30: Building AI Agents and Funding The chapter discusses the complexities and rapid pace of building AI agents, particularly in the dynamic environment of San Francisco. It emphasizes the importance of staying informed about AI developments, as missing out can lead to feeling left behind in this fast-evolving field. The chapter aims to convey essential knowledge on AI, especially for those not regularly following AI news, to help them catch up and actively participate in building AI agents.
01:30 - 02:00: Human Habits and AI Integration The chapter discusses the challenges of integrating AI into established human habits. Despite the availability of alternatives like chat-based AI and platforms such as Perplexity, Google remains dominant, highlighting the difficulty of changing user behavior. Founders of personal AI companies are tackling this problem by developing solutions that integrate AI into existing user environments, such as productivity tools like Notion and Obsidian, in an attempt to minimize fragmentation and make AI adoption more seamless.
02:00 - 02:30: Introducing AMX - Invisible Assistant The chapter introduces AMX, an 'invisible assistant' designed to integrate seamlessly with existing tools to help users manage multiple digital environments, such as WhatsApp and Gmail. AMX creates a personalized information retrieval and memory aid system by forming a layer around the user and organizing various types of data, including notes from Google Docs, SharePoint, and meetings held on platforms like Google Meet and Zoom. This centralized system aims to consolidate all data into a single, accessible location.
02:30 - 03:00: Unified Search Engine and Meeting Notes The chapter introduces the concept of a unified search engine that allows seamless interaction with AI without physical contact. Users can access and chat with their database across platforms such as Gmail and Outlook. This tool combines various data sources into one accessible point, allowing users to engage with emails and notes. The discussion hints at the ability to explore further how this search engine integrates different data, including emails and data stored in applications like Notion. This unified approach aims to streamline personal data management and enhance productivity by allowing users to easily access and manage their information from multiple platforms.
03:00 - 03:30: Interview with Sansar and Aon In the chapter titled 'Interview with Sansar and Aon,' the discussion revolves around a new tool designed to assist with capturing and organizing meeting notes. This tool operates under an 'invisible and omnipresent' philosophy, aimed at helping users without disrupting their existing workflow. It functions in the background to automate note-taking and enhances context understanding for professional meetings. Furthermore, this tool aids users in organizing their emails effectively, which serves as the first step in streamlining overall organization.
03:30 - 04:00: Ancar's Background The chapter introduces Sansar and Aon, who serve as CEO and CTO of a personal AI company. Their mission is to develop a personalization layer of AGI, which they describe as an invisible assistant that seamlessly integrates into existing workflows. The chapter focuses on Ancar's background, mentioning his upbringing in New Delhi, his undergraduate studies at DTU, and his subsequent move to London for work.
04:00 - 04:30: Arson's Background The narrator shares their background, mentioning their work as a software engineer at Bloomberg in London. They were also involved in maintaining an open-source project named Robin. After spending about three and a half years in the UK, they left their job to start their personal company, joining an accelerator called EF, which is a combination of an incubator and accelerator. This is where the narrator met Arson. The chapter introduces Arson, who is 19 years old and was born and raised in Kazakhstan.
04:30 - 05:00: Journey to Silicon Valley The chapter titled 'Journey to Silicon Valley' describes the early entrepreneurial experiences of the narrator. Starting a first company at the age of 14, the narrator quickly advanced to lead a machine learning engineering team at a state-based startup in Kazakhstan by the age of 16. This startup focused on developing a platform to assist high school students in successfully gaining university admission by mapping their journey from initial steps to achieving their educational goals.
05:00 - 05:30: Building a Web Framework in College In this chapter, the speaker shares their journey of working with various startups, spending time in Silicon Valley, and eventually returning to Kazakhstan to join Entrepreneur First (EF). During college, they built impressive projects, including an open-source web framework claimed to be better and faster than Django and Flask, which gained significant traction with 3 million downloads and numerous GitHub stars. The chapter concludes with a mention of starting their first company, hinting at more entrepreneurial endeavors.
05:30 - 06:00: Freelancing Journey The chapter titled 'Freelancing Journey' discusses the speaker's early experiences in the tech industry. Initially, they were part of a web agency that developed websites for companies during the pandemic lockdown. At the age of 15, the speaker started a startup which was acquired a year later when they were 16. Subsequently, they were promoted to the head of a startup following the acquisition. The dialogue also reveals ages of people involved, indicating the speaker is currently 19 while another individual is 24.
06:00 - 06:30: Open Source Contributions This chapter discusses the nature of open source contributions, starting with the experience of building projects in Silicon Valley. It highlights how having a good reputation and a successful track record are advantageous for obtaining funding. The conversation shifts to an individual's personal journey, noting that the speaker, named Robin, built a company that successfully raised $3 million. Additionally, it is revealed that Robin's entrepreneurial journey began at the young age of 14 with a company that built websites for other businesses, marking the first income generated by Robin’s first startup. The specific amount of money made by this venture is not disclosed in the transcript provided.
06:30 - 07:00: Arson's Educational Background and Startup Experience The chapter discusses Arson's early experience in building websites for clients globally at a low cost, using both WordPress and custom development from scratch. Additionally, it touches on Arson's lengthy background as a developer, starting from the age of 12, and mentions his academic experience during college in India.
07:00 - 07:30: Raising Funds with a Successful Track Record The chapter discusses the author's journey in software development during their university years. They were frustrated with Flask's lack of asynchronous support, especially given their positive experiences with JavaScript and React Native's async features. Inspired by the rising popularity of Rust and its use in the Deno project, the author decided to create a Python web framework with async support, written in Rust. They mention conducting their first benchmark, marking a significant milestone in this endeavor.
07:30 - 08:00: Finding Co-founders and Startups in the Bay Area The speaker describes the speed and efficiency of their programming project, specifically focusing on their use of Rust and comparison to C++. They explain how they improved the traditional Python web framework architecture by integrating the framework and server components, such as Flask and FastAPI with uvicorn or gunicorn, into a single entity, thus breaking conventional implementation patterns.
08:00 - 08:30: Challenges of Raising Funds This chapter discusses the innovative approach of combining a server and framework into a 'two-in-one' package, specifically in a Python web framework. This integration allows for faster data upload and processing, as opposed to using separate servers and frameworks like Flask and Rust. The new method is at least three times faster, showing a significant improvement in efficiency by reducing fragmentation and enhancing performance.
08:30 - 09:00: Economics of Building an AI Company This chapter discusses the background and early career of an individual involved in building AI companies. It begins with their first experience of earning a paycheck at the age of 17 through the Google Summer of Code program. The chapter continues by describing their career trajectory, including freelancing and serving as a CTO at a startup during their university years.
09:00 - 09:30: Funding and Growth Strategy The chapter titled 'Funding and Growth Strategy' begins by discussing the speaker's background as someone who has worked in a small startup that helped agencies with investments. This experience is followed by internships and a full-time position at Bloom. At the age of 17, the speaker earned $5,500 through Google's Summer of Code, which was seen as a remarkable achievement given their extensive contributions to open source prior to that age. Despite their experience, the speaker notes that open source work is not just about contributing code, but involves various other challenges as well.
09:30 - 10:00: Technical Details of Building an AI Assistant This chapter discusses the importance of community interaction and the art of giving and receiving feedback, particularly at a young age. At the age of 17, the narrator learned valuable lessons about overcoming arrogance and accepting feedback graciously, which proved to be one of the best experiences according to them. This newfound knowledge and experience paved the way for their freelancing career, working with startups and building websites for various clients. The focus is on the personal growth experienced through engaging with the community and the practical experience gained through freelancing.
10:00 - 10:30: Comparison of Technologies and Tools This chapter explores the journey of a young freelancer who specialized in creating MVPs (Minimum Viable Products) and small-scale websites for clients. The individual started freelancing through word of mouth while still only 18 years old. Despite not having a formal launch for their freelance services, they successfully gained clients through reputation alone. However, the financial rewards were not significantly high.
10:30 - 11:00: The Open Source Philosophy The chapter titled "The Open Source Philosophy" covers the experience of an individual who, after initially earning a substantial income through freelancing (around three to four lakhs or approximately $3,000 per month), found themselves disenchanted with the pursuit of money. This person describes their approach as not chasing money but rather focusing on solving problems—either their own or others'. Money is viewed merely as a tool to address these issues rather than an end goal. This mindset aligns with open source philosophy, emphasizing problem-solving, community benefit, and sustainability over profit.
11:00 - 11:30: Choosing the Right Models and Tools in AI The chapter discusses the process of creating an AI assistant based on the philosophy that humans share common problems, implying that solving one's problem might help solve many others'. The approach initially drew excitement through freelancing but soon became monotonous due to repetitive issues. Despite early financial success, generating about $4,000 at 18, the endeavor lacked sustained excitement and engagement.
11:30 - 12:00: Advice for Aspiring AI Entrepreneurs The chapter titled 'Advice for Aspiring AI Entrepreneurs' chronicles the speaker's journey of re-engagement with open source projects at the age of 19, particularly with React Native. The speaker actively participated in community interactions and gave talks at conferences. However, with the onset of the COVID-19 pandemic, there was a significant slowdown in activities, leading to a focus on academia and continued learning and practicing. This shift underscores the importance of adapting and persevering through global disruptions while maintaining engagement with the community.

"I Moved from India to London to San Francisco to Build THIS"! (2 Years of AI Learnings) Transcription

00:00 - 00:30 how much code is AI generated in your code base I would a lot argue 80 to 90% of the code generated what do you know about building an AI company now which you did not when you get started so people can figure out this is the right way to start so initially we thought it would be hard right but turns out it's actually harder than it's completely open source and what do you think about open sourcing an AI company where majority of the AI agents are not open sourced why what is the thought
00:30 - 01:00 process uh do so this is Jarvis a personal AI assistant AI agent which is built in less than 3 months by a 19yearold and a 24y old founder who moved from India to UK to San Francisco and from kazakistan to UK to San Francisco and how without a single idea with just with some experience you can get funded how to get that Jacob to get
01:00 - 01:30 funded what it is like building things so fast in San Francisco we'll get to learn all of that their experience of dealing with AI and especially building AI agents in the last 3 months all that knowledge is packed in the next 50 minutes which you can utilize and catch up because most of the kids right now who are not reading AI news every day they feel so left behind and I'm here to help you catch up so you can join the race and build AI agents as soon as as
01:30 - 02:00 possible so hello everyone human habits are really hard to change and integrating AI to your human habits has been challenging Google is still dominating even though Alternatives like chat gbt perplexity is there so we have founders of personal AI company who came up with the idea to go where users are already and integrate AI in the best possible way so the currently the problem is there's a lot of fragmentation you have tools like notion obsidian you have information in
02:00 - 02:30 WhatsApp Gmail even Gmail has multiple inboxes a person has to manage right so we created an invisible assistant called AMX that integrates into your existing tooling creates a layer around you and then helps you retrieve information and memorize things according to you so it shapes itself around so imagine your notes in Google Docs SharePoint Microsoft apps Google apps your notion all your data even your meetings in Google meet and zoom meeting all data connecting going to one place where you
02:30 - 03:00 can chat with the AI and you don't have to touch anything you just have the brain available at every single place you are whether using Gmail Outlook everywhere you go you have this database off your mind and you can chat with your emails you can chat with your notion data at one place this is fascinating can you go deeper into it yes so we created this unified search engine for you to help with the notes that you have taken already and once you use that you can search across all your online second
03:00 - 03:30 brain as we call it and once you have helped with that we help you capture the to notes where you don't want to take notes which is in the meeting so we follow this invisible and omnipresent philosophy where we don't want to interfere in your tooling but we exist in the background so we help you take meeting notes and gather more context about you but it happens in the background and once we have helped you with that the executive assistant also helps you organize it so we after that the first step for organization is is helping you organize your emails so hi
03:30 - 04:00 sansar and Aon can you please introduce yourself CEO and CTO yes that's correct um we are the CEO and CTO of the personal AI company and we are trying to build the personalization layer of AGI as we call it internally but what we mean by that it's an invisible assistant that quietly integrates into your existing workflow so hi my name is ancar born and raised in New Delhi stayed there did my undergrad from DTU then moved to London to work there then while
04:00 - 04:30 I was in London working as a software engineer at Bloomberg I was also maintaining the open source project called Robin and I worked and stayed in UK for about 3 three and a half years and then finally left my job to start the personal a company and we joined an accelerator called EF which is where I met arson it's like an incubator or accelerator both both so this is an incubator and accelerator where you got funded and you can work from yes excellent how about you arson I'm 19 I've was born and raised in Kazakhstan
04:30 - 05:00 uh started my first company when I was 14 LED an ml engineering team when I was uh 16 at a Siege State startup back in Kazakhstan at 16 yeah what was the startup it was back home uh we were building uh a platform for uh high school students who were helping them to get acceptence universities we were building a model like build that Ro map uh how to get from like from zero to one to to to University
05:00 - 05:30 uh after that I was working in different startups uh moved to Silicon Valley spent some time here then went back to Kazakhstan and get accepted to EF so you both have been building really cool stuff from your college days your app you open sourced a web framework better than Django better than flask and faster than it and got 3 million downloads and so many stars on GitHub and what what about you like what have you been building like when you break them down no actually started my first company at
05:30 - 06:00 14 uh it was not a start we were just building uh it was a web agency we were building uh websites for companies uhuh uh it was during the uh the pandemic uh during the lockdown after that I started the startup when I was 15 and yes got acquired when I when I was 16 and that's why I was promoted head of at a startup that got us acquired wow you are 18 now and you're 24 I'm 24 yes and I'm 19 19 really this is so brilliant so you both
06:00 - 06:30 had experience building stuff in Silicon Valley what I've seen is without idea you can get funding as well if you have previous success rates and reputation and your reputation of like building stuff that good Robin you built that and like got $3 million how did you make your first income using your first company at 14 yeah the very first company was we were just building websites for the company so it was it was my very first money that I uh uh that I made how much money you made through your first startup it was it was
06:30 - 07:00 like overall couple of thousands of dollars probably so building websites for clients all over the world for a cheap price like WordPress websites or like from scratch everything yeah we build uh a lot of Wordpress and a lot of websites from scratch okay and how about you can you break down the framework how did you build in your in your college in India so I had been a a developer since I was 12 so I had been building on different uh Technologies and when I was in the final
07:00 - 07:30 year of my University I was really annoyed that flask didn't have async support and I had been contributing to react native before that and I really like javascript's asnc feature so I wanted to create flask with it asnc support M but then rust was gaining popularity because ryal launched Dino and I wanted to learn Dino uh I wanted to learn rust so I thought let's create a python web framework with Asing support and let's write it in Rust and I I did my first Benchmark and it was
07:30 - 08:00 faster than anything else which is why I start yeah and it was fast because you went deeper layer in Rust as well as deeper layer of cc++ right just just rust so we how usually a python web framework works is they have a framework and a server which are different entities so with you have flask fast API then you have uvicorn or gunicorn which is a server so I combined both of them I broke the traditional uh implementation
08:00 - 08:30 of python and then tell told everyone it's a twoin one package wow so twoin one as in as in like the server and framework are shipped together in a python uh web framework simple example of this would be let's say I want to upload data from a website it goes to a different server and uses a different framework now with your two-in-one approach instead of using flask and maybe rust or any server separately you combine in two and one and made it faster as well at least 3x faster you
08:30 - 09:00 mentioned I think at least five times wow that is fascinating how old were you when you made your first paycheck through building stuff I think when I was 17 first year of University I got selected for Google summer of code which is how I got my first paycheck and later how did it continue the streak streak then I did freelancing for for a while I was a CTO at another startup uh in the third year of my University it
09:00 - 09:30 was a small startup for helping agencies Investments and then after that I have been employed through internships or through my full-time job at Bloom so at 17 you made like $5,500 to Google's summer of code it must be a like a joke for you to get in right you had so much open source contributions before 17 I saw your GitHub yes so it it was yes still slightly challenging because open source is not just just about contributing but
09:30 - 10:00 also finding a community interacting with the community learning how to give feedback especially learning how to take feedback at 17 because at 17 years old you are slightly arrogant you have uh you do not take uh feedback super well so yes it was still challenging but taught me a lot I would say one of the best experiences I've ever had okay and then you started freelancing for startups freelancing for clients people who were asking for websites startups
10:00 - 10:30 who wanted an MVP for them uh so just building small scale websites and dashboards for the while how was your freelancing journey go like where did you get clients how did you start building mostly through the word of mouth every everyone just knew there's this 18yar old who can create uh Prett websites and I actually had my first client before I sort of officially launched freelancing really fascinating so how much money you made through freelancing not that that much about
10:30 - 11:00 three to four lakhs then I actually got bored of freelancing in which it first month or something yes in two three months two 3 months you were making three to four lakhs that is like almost $3,000 plus yes and you were you were bored you you were not chasing money at the time so I never actually really chased money all my problems all my projects have been emerged from solving my own problems or solving a problem money or Revenue has been treated as a tool to solve this problem even here we
11:00 - 11:30 are creating an assistant that solves like our personal problem and then we base it on a philosophy that humans are more alike than unalike so if I am facing the problem arson is facing a similar problem probably you are facing a similar problem I see so idea is just you wanted to like solve problems and freelancing seemed exciting at first and then gave up because it wasn't exciting the problems were repetitive and not so at 18 you made like $4,000 and after
11:30 - 12:00 that what was the next gig you followed so at 19 I got back into contributing to open source again a lot uh was contributing to react native was giving talks at conferences trying to interact with Community then unfortunately Co happened so for one year I was not doing a lot I was uh involved in my Academia the world was slightly messed up as you remember so of contributing to just open source and practicing and learning more
12:00 - 12:30 things until I finally created Robin we're doing some interviews for companies uh created Robin then moved to London and worked at Bloomberg and maintained Robin the key thing to learn is without any monetary incentives you wanted to maintain that open source Library Robin faster than fast than Django and what what incentive you had to maintain an open source Library without even thinking of getting funded because in Silicon Valley raise funds are so easy if you have success like
12:30 - 13:00 Robin like 3 million downloads like open SS framework I think it was mostly because I needed something I really loved python uh and python by default was is slow if I could make it faster than anything else I could use it to make other things so always has been solving my own problems and just realizing everyone faces similar problems fascinating how about you uh so you made first income at 14 how much money you made uh 5K and then later on in your journey
13:00 - 13:30 10th grade at that time at high school and we're preparing for applying to universities and we had a problem where you don't know your chance of getting accepted you just prepare you take sat in different tests and uh it kind of a gambling you just apply it to 10 or 20 universities and see who who accepts you and we tried to build uh at that time I went through summer school at Stanford uh on data science that was my first time when I got
13:30 - 14:00 intrance of data science um and just got super interested in it to start reading a lot of books and U watching a lot of videos trying trying to build different models um and then we started that company with my high school friends uh classmates we built an algorithm that would uh predict your chances of getting accepted based on your test results on your essays on your uh basically your student profile everything have and
14:00 - 14:30 everything that you're going to send to universities So based on that we're giving you chances of getting accepted uh yeah and at that moment a year later in 2021 uh company um seat stage startup in kazaks and they were evaluated 8 million at the time they acquired us uh and it was kind of they aquired as basically they just got that algorithm and they wanted to integrate it into their so hired plus hired it took your people so
14:30 - 15:00 that's AC hired huh and I was leading the ml engineering team at that time uh I was integrating that model uh that algorithm into their product um and I was just getting uh a lot of experience of running an actual uh startup with with Revenue that was my first time working at at a serious company I'd say wow so you were building ml models and you got acquir because machine learning was a big hard thing so
15:00 - 15:30 what kind of models you were building like at the age of 15 16 oh a lot of I mean different models I was just exploring um yeah obviously started from from very basic statistic statistical models at Stanford wow you were at Stanford at 15 yeah that was a summer school yeah okay so Stanford summer school you were 15 huh uh I'm not sure if I was 15 but yeah that was a while ago like I remember how old that was but yes uh that was first
15:30 - 16:00 time when I uh kind of uh understood what ml was in the time because I tried to read a couple some books before the summer school it was very hard to understand what what's going on uh I mean I did some calculates before that but still didn't get it um so yeah everything my ml Journey sorry from Stanford you both had great success rates in the past and still how hard it is with that much success rate
16:00 - 16:30 like creating a company getting acquired or equired and then building a great open- source project so how easy it is to raise funds with that success I I think it's still quite hard and funding is not just about your reputation it's also not just about the past it's also about the future right so you still want to focus on your product find a product that people ask actually want and then you want to inject the
16:30 - 17:00 funds into growing it or creating the best type of product yes reputation helps but still going through the same same Rocky journey is what it takes to build the most polished um yeah first I totally agree with that it's not just about your background it's about the metrics like the actual retention how many users you have and how many users of them are um using your product on a daily weekly monthly basis so yeah totally agree with that so how did you guys meet like so is it easy for any
17:00 - 17:30 18-year-old coming to Bay Area and finding co-founders and get raising funds with that success rate you would say no that was very hard really really really hard to be honest yeah it was very hard for me to get a Visa here firstly for the very first time uh I came here on B1 Visa spent some time here just exploring working with different people thankfully the we have a huge community of Central Asian FS here uh so I got a lot of support uhhuh um but again still wasn't able to build
17:30 - 18:00 something truly successful and something I I wasn't able to find an idea that I would really really love um because I before moving to the Bay Area I I was working at a startup as a chief of operations um and that was a very Niche B2B startup for HR and after that I tried to build a lot of B2B specific Nar Focus ideas to solve some problems for
18:00 - 18:30 businesses and at some point I just realized that I don't want to do that I want to build something like for myself uh something in the consumer Market something just something cool because it's just the perfect time for that and I'm young and I have all the resources to that 100% like B2B startup space is extraordinary like I would say like 90% of the startups in Silicon Valley are B2B B2B s because it's so easy and fast to get profitable like Merk or AI 100 million AR in less than one year for
18:30 - 19:00 example deal rep deal got 800 million AR that fast so it's fascinating everyone wants to build in B2B space and like you guys getting into consumer Market d2c direct to consumer or b2c whatever you want to call it it's just fascinating and you were almost almost close to building something in B2B SAS what what made you give up where the success rate is highest uh I think yeah first I picked an idea that was uh in a very very crowed market and I uh what was the
19:00 - 19:30 idea I was building um tool for uh first it started as a storage for llm prompts that was like the the initial idea just store your prompts uh you didn't need to change them in your production code you just sort them somewhere outside of your code so it will be very easier to erase with prompts uhuh and yeah I couldn't find um turned out that that was not that big of a problem for companies um and at some point I again realized that
19:30 - 20:00 too many B2B narrow Focus startups and now companies have to manage three 400 different tools and as a person as a consumer you still have tens dozens of different products that you have uh to use you're forced to do that you can't like at this point you can't change that you just have a lot of tools and you you're forced to use all of them 100% now let's talk about the ecosystem in Europe you had to come from from like UK you were working at
20:00 - 20:30 Bloomberg yes why did you put your job at Bloomberg to come to B area that's a great story by the way uh so I was really not excited not feeling the excitement yes the people were nice the problem was good but after a while everything gets repetitive uh I was working at Robin so I was seeing the direct impact that people got through my abilities through the tool that I created and then and I had an idea of
20:30 - 21:00 building Amir and the impact that you could have through that direct to Consumers is what excited me at Bloomberg it we were serving like businesses directly which is clearly not something I'm really excited about so just serving the people and serving myself so you got bold working at Bloomberg your idea was that you wanted to pick an exciting problem how did you guys come up with this idea actually was working on this idea on the side for the past 9 months before me meeting arson I
21:00 - 21:30 tried different approaches tried something in Hardware tried a desktop application but this idea had been building in in my mind for a while did a lot of experiments and ultimately I decided to qu Bloomberg go full-time on the startup wow call call my friends I knew some people at entrepreneur first join entrepreneur first and met Haron there so fail fast as a principle like you know Amazon has the story of Auto a ating building a startup like they
21:30 - 22:00 created a search engine they created a phone to building automating testing failing and then quitting the project and then trying something new building starting failing and then starting something new so did you follow this structure when you were trying different startup ideas 100% actually like I usually don't want to start a startup I have an idea and just realized that starting a company is the most sustainable way to go I tried different approaches for ideas like I was
22:00 - 22:30 contributing to have impact I even created Robin right but at some point if you're doing purely open source it burns you out because you have eight hours of job during the day then 5 hours working on a side project then managing everything your personal life after that you only have 24 hours in a day so I really wanted to create this idea to a reality that's why we started a startup but us being technical is the biggest advantage that we can ship fast fail first I trade first how did you guys
22:30 - 23:00 meet then like so you had the idea already and how easy it's for someone like with 18y old or like 18 year 19 year old like you or like in 20s to come to Bay Area with idea without idea and just get funded so like first you can talk about how do you guys meet and then how easy it is okay we met in London actually in entrepreneurs first so this has a like entrepreneurs first is in UK as well as San Francisco sorry it starts so it has three three ways there's a
23:00 - 23:30 cohort in San Francisco as well but we met in the London cohort we got funded there and then we came to SF through after being funded after raising a preed oh so so you guys met there and so you first moved to UK as well from Kazakhstan I didn't move there I just got accepted to the program oh nice you applied remotely to Startup exchange no what it's called entrepreneur so you applied directly to entrepreneurs first you applied different accelerators incubators and you got into this one yes
23:30 - 24:00 so like currently with extremely tough Visa conditions you think it was easier because you got into UK like right now it's like Visa is getting really hard for everyone like you first had to get into UK and then getting transferred to SF that was the process um sort of yeah it was a lot easier to get a UK tourist visa uh but then when we got funded in January I had to go back to Esta to get a US tourist visa oh you got rejected once and then I reapplied next week and
24:00 - 24:30 got accepted and then went straight away here wow so like yeah because this is a really important concern most people are scared getting Visa like these deportations and like visa restrictions what what are your thoughts on that like right now us has always historically favored people building a company contributing to the economy directly yeah so I don't think that's changing anytime soon you're you're contributing directly to the economy right to the they want as many corporations startups
24:30 - 25:00 as possible if you're if you're successful you're creating employability in the future yeah 100% and you're paying a lot of taxes amazing how many accelerators incubators you had to apply to finally get in like right now when people talk about getting funded right you taught me that every incubator every accelerator or every VC want a J curve right yeah yeah they just want like oh you're growing right now if you
25:00 - 25:30 have hit the Jacob the chances of getting funded is the highest you have little bit of exponential growth you have the highest so how easy it is when you have that Jacob applying to different incubators VCS and then getting in so entrepreneurs first is slightly different here they invest in people not the ideas they actually I was the only person who went there with an idea they usually encourage you to come without an idea wow they invest in the person they talk to them understand
25:30 - 26:00 their motivations and then pair people they basically invite a lot of smart people and hope that you'll find a smart person who you Vibe well with and would want to build a company with I see this is a fascinating idea but like it's not easy to get in but your previous track record helps you right yes okay so you shared the track record now let's talk about the company how do you build such a powerful AI assistant which goes wherever you go and you can search in it so what it takes because right now there
26:00 - 26:30 are so many cool Technologies coming in like mCP so let's talk let's start with that so firstly the idea is quite contrary to what everyone recommends you there's been a classic wisdom of doing just one thing but and doing it well right but we created this layer that integrates a lot of existing tools so first to convince and convey that idea very well is a hard part but then after that technically we created one of the
26:30 - 27:00 fastest retrieval algorithms that we use for AMX it's I think 500 times faster than Lang chin yeah it's actually more than 500 more than 500 times faster than Lang chin which is used for the retrieval uh of C really so like Lang chain people building AI agents connecting different AI apps and like like let's say connecting Gmail to Google Docs like that Lang chain connecting different apps and AI application is f 500x slower than your approach yes so how did you build such a
27:00 - 27:30 powerful connector that's a trade secret technically yeah technically that's a trade secret but the the idea came from like arson was already working in the retrieval space because of his previous startup and I proposed him the idea that we can build an AI assistant that follows you everywhere so it started with meetings then we created the search engine and that was just the perfect technology to
27:30 - 28:00 integrate so we worked together refined it and it like it became an evolving TP that scales well with your document wow so now you can connect like different apps Gmail Google Docs 500 T faster and the data flows through to the AI chat like wow I know trade secret you cannot share the algorithm but can you share the agentic flow how can someone out there build such an agentic flow so even before a gentic flow our focus
28:00 - 28:30 is to get contextual understanding about you people are building agents right now but I believe no one's really using a lot of Agents just because they're not smart enough so we are focused on making the assistant smart enough then possibly doing some agentic task for you so the first the most important challenge according to us is contextual understanding about a person for an assistant to work well if you were to
28:30 - 29:00 hire an executive assistant first you would expect them to know about you without before them telling you any suggestions right so to understand a user how would you do it you would extract all the emails and understand the style understand that data or understand all the nodes which is like a lot of data how do you understand a user that's exactly what we do we form this memory layer around you and the memory layer you can to build a memory layer we help you in a certain way so we allow you the fastest access and then capture
29:00 - 29:30 notes for you where you don't want to take so we capture meeting notes for you so that you don't have to take any but for the notes that you've taken already we have this unified search engine across your second gr oh do you use like a database or do you create your own postest database for memory layer like like rag system Vector database or did you like use some existing ones like m zero so we created we used a super base but the retrieval algorithm that we use over the top is uh sort of trade
29:30 - 30:00 secret okay fascinating and here lot of models come into picture can you share what understanding you have with models right now and how did you leverage that understanding while building it yeah that's actually the reason why we are going horizontal first because if you go vertical if you focus on one Niche for example emails you have Contex just about the emails um or if you go to meetings only you have the context only by the meetings and probably only about some like one job that they do but since
30:00 - 30:30 we're going to Arizona we can get as much data as possible now there is available in the software like different tools like Gmail Google Docs notion that we will be expanding to other tools uh so that's why we're going horizontal and yeah we just we just find a way to Reeve to to sear that information very very quickly and use it in a in a way that you would need it so it won't distract you but it will help you when you need it okay so you guys are going horizontal
30:30 - 31:00 on different apps because I see so many competitors of product hunt which one one app that categorize your Gmail well and like you know but you're going every single you not just cleaning Gmail but you know connecting all the apps but what model leverage you use like what models you use how do you leverage it well yeah we for TCH generation we basically use LMA 3.3 uh almost everywhere I think everywhere yes but we use open AI as well so we don't have any
31:00 - 31:30 model loyalty we we will use whatever is the best model we were considering Gemini for for for the for one of our use cases internally today so as the model aay incre improves we actually benefit a lot from it so we don't have any model loyalty whatever is best for the job we use it in different places with better models keep coming like you know for example chat GPT released a GPT 4.5 which is extremely expensive how do you leverage the cost and and the usefulness of a model in the right place
31:30 - 32:00 so we in in addition to the usefulness like the smartness we also have a criteria which is the faster inference so we balance them with like the fastest inference with the smartest intelligence in addition to the cost so fastest inference so what model right now you see is fastest I see mistol really fast these days yeah it doesn't it does depend on the model but it also depends on INF infrastructure which is uh for now we use Gro mostly because they provide one the fastest references available and affordable but Gro three
32:00 - 32:30 or two no Gro not the Twitter one but Gro with q in the end it's the company that built uh gpus specifically not for training but for influencing models oh so you use their uh infrastructure and where you can use any model you want and inference it that's the idea it's not any it's like some models they provide uh for their infrastructure but we use llama 3.3 there um which is a 70 billion
32:30 - 33:00 model for entrance why Gro over open router or stuff like that this groc has the fastest INF really so it provides the fastest in it's it's like really growing which where is it from the gro company uh s they're here in that view oh wow so like they're providing fastest inance because they have like more relationship with how how no they built we actually went to to we were demoing under HQ last week I think and they kind of shared why they're faster than others
33:00 - 33:30 and the reason is that if a lot of companies are using gpus and inferencing their models on gpus but they came up with uh an infrastructure that allows that it's focused only this designed just for influencing but not for training because gpus initially was not for training models at all it was just for playing games and people realize that we could use it for training models and gr figured out a way to uh to build something that that allows you to influence models to run them faster
33:30 - 34:00 fascinating that's amazing so like using the right tools at right places that's an art and you guys understood by being in Baya it all connects if you had to build this because you got into entrepreneurs first the incubator how would have been how would would it have been different building in UK versus us or you had to come to San Francisco so we built the meeting co-pilot in in UK but we expanded we started expanding in
34:00 - 34:30 the last week where while we were in the UK we are here in SF just based just feeding of the energy people are excited about the AI assistant uh space people are excited people are earliest adopters here we have been interacting with people we have been interacting with you just the excitement and the technical aptitude is Miles Ahead of any other place yeah I think the it's not just the ecosystem what I've noticed it's the willingness to try here whether
34:30 - 35:00 you go to product hunt whether you go to any conference people want to try new products people are more welcoming change and other other places all over the world maybe they have the best startup but the idea of welcoming change is really hard and you welcome change when it's right in front of you somehow or the other so that accelerates your growth you got thousands of users right now how did you get your first thousand users so we posted a lot on Reddit actually oh we went viral on Haka news
35:00 - 35:30 we were on the first page of Haka news twice through Reddit and through demoing at various places in SF we were at the Grog demo we were at a couple of other demos I think SF here has the best people who will be your earliest adopters because they share the same mentality okay so this is brilliant wow so it's like you started sharing as many places as possible without so without invest in ads you started sharing as many places as possible and through word
35:30 - 36:00 of mouth and presenting in person in SF that's the most valuable thing 100% of our user her organic where you never spend a dollar on marketing brilliant and here what do you think about scaling like you have raised your first fund how you plan to use it the preed fund and maybe future seed fund to grow as much as possible I see the applications it's so cool how do you see see growth so the growth has been thankfully very very nice due to the word of mouth and we
36:00 - 36:30 want to leverage that for as long as possible people invest in ads too early and I I believe that's them fooling themselves they inject money into just raising the numbers but all of it has been organic for us not only that will help us verify the product Market fit but also know tell us when it's the right time to invest in a proper growth push so where do you plan to invest the funds you ra a lot in in infra hiring the best people creating the best product possible uhuh getting designers
36:30 - 37:00 getting engineers and then doing a growth push eventually what is the most expensive part of building an AI company in 2025 because right now people with two to three how many people are in your team two just two just two of you how what is the most expensive part I think depends there are two kinds of AI companies we are an application lir company for us it will still be Manpower MH and and if when it comes a growth push but for mod level companies at to the same
37:00 - 37:30 so Gro and everything it's cheap so if if I if I have to consider one API call how much sents would it cost like you know like one user per 30 minutes of usage how much it costs you if you have to break down Gro I think it's very cheap probably cents maybe 10 of cents 10 in AI there's a new Mor law 7 months AI capacity cost capacity is doubling cost is going half so with that it's really cheap but it's
37:30 - 38:00 just the Manpower and two of you have built it end to end and now you can carried forward why do you need more people then just because we have a lot of users so that we can ship even faster uhhuh it's not we we are not targeting a very fat company where we hire everyone immediately we still want to uh keep a lean model but the highest quality Engineers usually require a very high pay rate yeah in AI company right now the biggest problem is you know the J
38:00 - 38:30 curve is really fast like people are willing to try out and the people usually you know first month they will subscribe and cancel next month that happens to so many a apps and within the first month Revenue people multiply by 12 and calculate the ARR so it can be faked a lot so that's why most investors wait for 3 months if the J curve is happening for months so how do you think about that how much growth is predictable and how much growth is just trying and then fall out I think it depends on the product we have been
38:30 - 39:00 Organics and we haven't been pushing our attention is quite High actually so we haven't been pushing a lot on ads so we know the users that who are coming sticking to us they are interacting with us in the community it's usually a bad sign if you are unable to talk to the users uhhuh that's usually a bad sign if they're not interacting in the community or they're turning immediately and just to fake the numbers you inject some money and get a jov and also I think speed of shipping is also what matters a lot um because we yeah we we have to
39:00 - 39:30 talk to a lot of our users and we understand what they what they want want we want to see in our product so we keep shipping new features very very quickly and we learn from that that allows us to keep the retention and grow uh make the product actually sticky the users are discovering a product through the word of mouth that means someone who really likes a product is telling their friends about this good they want to use so I it's not artificial ads driven growth
39:30 - 40:00 they try and forget it's it's organic growth so that's how you understand there's actual demand for it and what do you know about building an AI company now which you did not when you get started so people can figure out this is the right way to start so initially we thought it would be hard right but turns out it's actually harder than I think that's the beauty about entrepreneurship like everyone thinks it's hard and when everyone starts to it it's even harder than what theyve imagined but what I
40:00 - 40:30 would still do is think very carefully about what I want to do and then uh just instead of just strapping AI over anything just really deeply think about the product and I'm willing to commit five to 10 years to it amazing awesome and let's go into the open source side of your project it's completely open source and what do you think about open sourcing an AI company where majority of the AI agent are not open sourced why what is the thought process I I grew up
40:30 - 41:00 reading Richard stalman and Linux towal and who are they so Richard stman is the person who created G like the Creator behind tools like emac and other core libraries for the Linux and lus tows is the person who created Linux and I owe and attribute a lot of My Success to contributing to open source software right I started contributing to open source software when I was 16 I started using Ubunto when I was 12 it feels
41:00 - 41:30 weird when you have taken so much value from the community and once you actually want to sustain create start creating value you just add a pay wall to it you had a add a layer of hiddenness to it so we want uh to be open again the web was open first when I started learning engineering it became closed by all the obfuscation just want to be open again uh when web was closed like so so initially when I started learning HTML
41:30 - 42:00 you could just inspect the source and you could see the entire website's code right now it's just a single file minified JavaScript minified HTML that's super hard to be about I see so like websites were more open source you could hack them on the front end even the back end sometimes by adding a script there exactly yeah you could just write SQL injections that was yeah yeah like right now also like lot of websites you can like you know write a function in the JavaScript and if a function is hidden you can do that like injection but it
42:00 - 42:30 it's really hard now as people are getting more secure so yeah so it was Clos it's closed now but you want to start with open see the get the feedback and contributions from all the developers in the world make it as better as possible and later on if needed you can make it close that's what saying no we'll probably never make it to close it's just against our models a because we value transparency and how people start believing that AG are predicting that AGI will be this powerful technology in the future right if it's that powerful it should be
42:30 - 43:00 accessible to everyone and given the fact that we are a layer over AGI we are a consumer space it makes no sense to and if we become a powerful company why would we want to be proprietary right it so just on models and values yeah that is brilliant way to think about it right now because even open I said they were on the right wrong side of the history by choosing close Source now I want to uh go into the depth of technicality so if someone wants to start building an
43:00 - 43:30 AI company right now what model approach would you recommend where to start if you had to build a road map connecting your dots backwards how would you build a road map for anyone watching I would still think about the problem that you're trying to solve for us it is solving all the mundane tasks in a human's life right and then see how you can leverage AI into that regard the the idea there is AI company or not AI company company it needs to solve a problem that people care about and then
43:30 - 44:00 use how then see how AI can help you be better for example if I'm trying to solve the search engine 10 years ago it wouldn't be possible because the technology wasn't good enough for us to make a unified search engine now the llms are so good we have the retrieval algorithms where we can actually do a unified search which has enabled us so probably the first thing is identifying a problem and then Pro connecting it to time seeing
44:00 - 44:30 what problems are solvable now compared to 10 years ago which were not solvable back then so think about the problems first yes what problems are solving now because we can have leverage with AI and think about the problems and like right now tools are infinite it's really easy to use any tool any AI agent can be connected with llama uh L Lang chain m CP if you have to consider mCP and Lang chain right now building AI Agents from
44:30 - 45:00 scratch what do you think is the future what do you think has the maximum use cases I honestly think agents will change a lot like even Lang chain or mCP is not the future maybe it will be the future will be created by the creators of those two companies but no one really knows what agents are everyone has a different definition so agents will evolve so well the tool mCP has has a good direction in that regard where they are trying to standardize the server but I still think
45:00 - 45:30 there's a lot of work to be put in in that regard really to to understand the AI Technologies everyone needs can you break down what technologies we need like I I know that Lang chain is needed mCP is maybe needed can you list all the Technologies breaking down we need building an AI company yeah agent agent the problem you're solving depends highly on the problem you're solving we can talk about the application layer company that we had trying yeah so an application layer company if you can
45:30 - 46:00 share all the AI agentic tools we need Lang chain MP can you list them down and break them down all right so depends so if you want to make an agent probably mCP is a correct ecosystem to follow right now because it's people are investing money in it but we focus more on the retrieval side to get contextual understanding and turns out power retrieval algorithm is faster than longchain wow so you don't
46:00 - 46:30 necessarily need anything yes there are certain top level topics like a rag is useful llms are useful but given enough R&D I think you can probably swap them out for for your own implementation wow so let's let's break them down first of all what is Lang chain what is mCP so Lang chain so Lang chain just pulls the data from any sort like Google talks to gmail which you're trying to bring but your algorithm is
46:30 - 47:00 500 faster yeah but what about MCB where does it play Define mCP to so mCP comes in when you have to interact with the agent somehow you want to give it some interaction basically the side effects of an agent you want it to be interactable with your computer or your browser so it's a standardized server implementation I see so Lang chain comes when you bring the data mCP comes when you interact with all the multiple data which you have pulled out when you want to take the actions on the data so I I
47:00 - 47:30 would say an mCP is a standard for connecting AI assistant to systems where the data lives so MCB model context protocol I would imagine like mCP being the HTTP of AI agents it's just a protocol people are creating servers for that but what um people are proposing mCP for is just a uh standard it's a protocol which GitHub is using it's a
47:30 - 48:00 standard so many companies are using GitHub is using with which you can connect and have an agent like I can just type in in my cursor AI create a project and push it in one command it would push it as well so it connects different apps but you still need Lang chain which pulls data so here Lang chain will come in Google Docs you to pull data from Google Docs you'll use Lang chain to pull it but your algorithm which you're using is 500 faster than Lang chain so that's how these two connect what other tools in AI agents application layer you think of you
48:00 - 48:30 obviously need D llms which you use on Gro gr Q not K so groc is the llm hosting infrastructure so llms could be uh divided into two parts one is the hosting infrastructure people can even sell for our application run it on their own laptops running on their own servers so for hosting infrastructure use Croc then comes the llm models which is open AI uh Facebook l Mells models and all the other big providers you definitely need those two
48:30 - 49:00 and then comes the traditional tooling like your front end tooling your backend tooling your frontend web Frameworks backend web Frameworks a database you can use a vector database or you can use a traditional SQL database and add a vector extension to that wow so the extension is now you have data coming in data agenting with different apps now the next goal is storing it and you store it with the vector database most people use postris as a vector database
49:00 - 49:30 and with connecting with the vector database you have now with postris a chat board you can chat with that data which you have so this is how the in to end app is made and how long did it take you all to build it like multiple application you have like connected with notion Google Docs Gmail sorry so we actually built the first version in a night really yeah the very first the very first version of the app was built
49:30 - 50:00 in a night but obviously to polish it and refine it we have been working on the application for three two months now uhhuh and that's all it took for us to create all the integration wow so two months all integration and you guys are building from the last 3 months now wow this is fascinating why do you all say this is the best time to build anything any startup for businesses consumers right now because the current AI tooling is so good you have GitHub copilot you have cursor that
50:00 - 50:30 can accelerate your workflow in such a great manner that you you don't need a big team you don't need like a devops person you don't need someone configuring your database handling it you don't need a designer because you have tools like VZ obviously you do need them eventually but if you lack if you have a skill Gap at a certain place the AI tools accelerate you and make your startup dream possible yeah what are the tools you can think of
50:30 - 51:00 that has helped you build it that fast within one day you had the first version would you say cursor AI VZ maybe bulge so what would you say what are the tools so we use cursor Ai and VZ a lot obviously with Chad GPT and we sometimes also use uh figma but mostly it's uh CH GPT cursor and VZ wow it's ch GPT cursor v0 what are the tricks you would recommend people to
51:00 - 51:30 think of prompting to build stuff overnight like you all uh yeah we have very high prompt engineering skills right I think the models have gotten so good that for the initial use cases The Prompt engineering skills are not the bar for prompt engineering is not super high you can just ask it to generate an app especially with tools like lovable and bolt but as you as you go further I don't think prompting can even help then comes the software engineering skills at
51:30 - 52:00 this point wow still the application fascinating so if prompting is not needed as much right now how much code is AI generated in your code base a lot I would argue 80 to 90% of the Cod yes so go ahead everyone go check out Amir from the personal AI company in the description below and thank you so much sansar and arson for giving incredible in incredible incredible insights and wish you wish you all the best thank you for having us yeah thank you