Tarun Raheja (@atemyipod) header
Tarun Raheja (@atemyipod) profile picture

AI/Robotics investing @accel | Prev - UPenn, MSFT Research, JPM Quant, @bitspilaniindia | Opinions my own, not employer's

San Francisco, CA
Joined March 2016
Photo by Tarun Raheja (@atemyipod)Photo by Tarun Raheja (@atemyipod)Photo by Tarun Raheja (@atemyipod)Photo by Tarun Raheja (@atemyipod)Photo by Tarun Raheja (@atemyipod)Photo by Tarun Raheja (@atemyipod)Photo by Tarun Raheja (@atemyipod)Photo by Tarun Raheja (@atemyipod)Photo by Tarun Raheja (@atemyipod)Photo by Tarun Raheja (@atemyipod)
It was lovely to host you guys, and looking forward to some extremely fruitful collaborations in the future!
This is obvious to most people on the ground but the AI founder + researcher talent in Bengaluru is really great. Hosted a dinner with @AccelIndia this evening with a small group and the range of topics was amazing, the vibes were great and I learned so much.
1
152
Tarun Raheja retweeted
3 interesting (and counter intuitive) takeaways from an analysis of Bay Area startups founded in the last 4 years that have already crossed a $500M+ valuation: 1. IIT is one of the top 5 feeder schools for Bay Area startups valued at $500M+! Others might take it for granted, but pause and think about that: continents away, in a completely unfamiliar market, founders who spent their first 20+ years in India are building at the very top. I find this extremely special - there’s no other school from any other country on the list. 2. 80% of these founders come from a Technical Background. The ratio is flipped in India today, though we are already seeing the shift happening. Expect far more Technical CEOs as AI and R&D-first companies in Manufacturing, DeepTech, and Defence take off. 3. Age doesn’t actually matter. Founders across different age brackets are starting up and scaling to $500M+ at similar rates. “Only 20-somethings get funded” is just a loud narrative! Thanks and well done @utkbajpai, keep ’em coming!
Replying to @utkbajpai
Another narrative violation: these aren't all 22-year-old dropouts. The median CEO was 33 at founding, a fifth were over 40, and only 14 (9%) started straight out of college. About 80% are technical. (5/8)
8
1
1
60
3,064
this is an insanely cracked team working on very exciting fundamental research! the winning formula for a moonshot is to have sharp founders working on a narrow thesis with deep focus. expecting to see insane things from these guys soon ✨
Some news: I'm starting a company in the AI for Science space. We are training foundation models and improving them for multi-agent scientific discovery. If you're looking to join an early-stage startup as a Research Scientist or Research Engineer, I'd love to chat. My DMs are open!
14
660
Tarun Raheja retweeted
For anyone curious how venture backed Indian companies founded in last decade (since 2015) have done: 14 are already listed, worth ~₹4.2 lakh crore (~$44B). Interesting thing to note: almost none of them are SaaS or global narrative plays. They're India-focused: consumption, manufacturing, financial services. No reason why the next decade (even if we are lagging the AI race) will not produce larger outcomes for India. We need to play to our strengths 🇮🇳✨
11
9
5
112
16,624
i follow rishi for golden insights like this btw. thoughtful dude
last gen consumer startups (uber, airbnb, doordash) won primarily by organizing fragmented real-world supply. consumer ai will be fought over organizing demand: ai helps capture the rich intent people struggle to articulate (constraints, soft preferences, trade-offs), express that intent on your behalf, and route it to existing supply.
2
1
114
26,945
last gen consumer startups (uber, airbnb, doordash) won primarily by organizing fragmented real-world supply. consumer ai will be fought over organizing demand: ai helps capture the rich intent people struggle to articulate (constraints, soft preferences, trade-offs), express that intent on your behalf, and route it to existing supply.
22
33
10
420
45,241
Tarun Raheja retweeted
At seed, an investor's conviction is sometimes just your confidence reflected back. Aggression gets you the meeting, sometimes even the cheque. It doesn't get you the company. On hollow aggression, why a round isn't validation, and the 6 questions:
25
9
5
143
15,045
India AI is booming, and my good friend @SaSarthaksingh @AccelIndia is at the centre of this! The opportunity is immense and unprecedented. Doing principled work at the frontier of economically relevant domains is central to building the future we envision at @Accel.
We are HIRING. Interestingly, few of the best companies being built in India right now might be the ones you've never heard of. Three Accel portfolio companies I work closely with have raised ~$150M+ between them in the last six months. We don't believe in big announcements or building in public. Just heads down. Every one of these founders have been telling me the same thing: talent is the bottleneck in India. So I'm helping with it myself. They're hiring across engineering, operations and generalist roles (EIRs, Chiefs of Staff). The bar is very high. Why I think they're among the best places to work in India: 1. AI at the core of the product 2. Global exposure, with US customers and markets 3. Growing 50%+ quarter on quarter 4. Founders for 2 were CXOs at large public companies 5. Just getting started, so the equity upside is real If you know someone who's the best engineer or operator on their team and are quietly bored, send this their way. If there's a fit, I will walk you through each company: what it does, how fast it's growing and why I think it wins. . Form in the reply 👇 or just DM me!
4
38
2,780
Today I'm releasing a deterministic algorithm that tells you whether the number of primes up to x is odd or even, in x^o(1) time. If x has n bits - Previous best: √x = 2^(n/2), exponential in n. Our algorithm: 2^(n/√log n). Paper: hexagonmath.org/pdf/2611.001…
4
11
1
92
4,399
Congratulations Dr. @dvsaisurya ! Rare picture of two top tier collaborators
Congrats @dvsaisurya on your successful defense! Its been a pleasure of my life to collaborate with you! @inderjit_ml @cho_jui_hsieh @Devvrit_Khatri
10
552
Tarun Raheja retweeted
The main thing is to keep the main thing, the main thing.
1
1
50
3,003
Tarun Raheja retweeted
I'm on the industry job market this year! I'm a PhD student at CMU LTI. My work spans post-training, evals, alignment and multimodality! Please reach out (DM) if you are hiring and think I'd be a good fit, I'd love to chat :) Website: simran-khanuja.github.io/ CV: simran-khanuja.github.io/ass…
11
69
3
485
41,580
Tarun Raheja retweeted
We asked @alma_inc to make an @a16z style video, without a human using the editor. Fully with computer-use. Alma merged the clips, found the characters, added name cards and dropped in transitions on its own using Adobe premiere pro. You no longer need to learn any software. Stay tuned!
29
20
17
173
39,923
Proudly designed & manufactured in India 🇮🇳
Today, we are proud to introduce… Mitts 🧤 State of the art tactile gloves that capture high-precision touch data for robots. Vision alone isn’t enough. Mitts record touch and movement with world-class precision, helping robots learn to work in the physical world.
73
177
8
2,361
55,089
Tarun Raheja retweeted
Today, we are proud to introduce… Mitts 🧤 State of the art tactile gloves that capture high-precision touch data for robots. Vision alone isn’t enough. Mitts record touch and movement with world-class precision, helping robots learn to work in the physical world.
55
156
40
1,336
111,576
Not surprised - Batman is an MLE
Ben Affleck reveals he writes Python, understands convolutional neural networks, worked extensively with GPUs, and used his celebrity status to get private looks at Google and OpenAI’s video models “I’ve always been kind of into computers since I was young. Then, when film started to move from analog film to digital, I became more interested in that aspect of it. The visual-effects workflow for many years has included machine learning, so I can write pretty shitty Python scripts and stuff like that. “With convolutional neural networks, which were the precursors to what the transformer can do, which is much more computation simultaneously, you would do things like look at what’s called a tensor. That’s the numerical translation of a visual image in numbers, like the batch number, the frame number and the red, green and blue values of each pixel in each frame. It’s just that simple. That numeric is called a tensor. “You’d use a convolutional neural network to identify patterns that reveal what’s called edge detection or feature extraction, which is identifying patterns well enough to know, this is where the window ledge is, so we can more easily take the green-screen image out and replace it with something. “That was familiar to me early on because, prior to Artists Equity, I had a small visual-effects company. I’ve worked with GPUs a lot too. The visual-effects guys said, ‘Hey, you should see. There are a couple: Google and this other company, OpenAI, are doing really interesting stuff with transformers in video.’ “I’ve learned that I can actually just call up and go, ‘Hey, it’s Ben Affleck. Can I come see what you’re doing?’ Sometimes people say yes, to my astonishment.”
1
7
421
Tarun Raheja retweeted
Are you building vertical AI products? This report offers several insights and examples. This helps you decide where to focus your efforts and how to make your engineering future-proof. From our resident expert, @atemyipod, who has made it concise and useful. Please download and share with others.
Introducing The Working Knowledge, a new technical publication from SeedToScale by Accel to help founders and technical teams make better decisions about the products they build. It brings together AI research and lessons from practitioners in playbooks teams can put to work. Our first, “How to Build a Vertical AI Product” by Tarun Raheja, explores how to evaluate models for your domain, build the context and tools around them, and make the product reliable in production. With a framework and worked example, it also helps teams decide what to keep, simplify or remove as models improve. Explore The Working Knowledge and download the playbook: seedtoscale.com/the-working-…
13
13
158
25,633
Wrote this piece on harness engineering to enable people to build solid AI products with @Seed2Scale @AccelIndia @Accel Would love to see more application layer companies built out of India, and to see people going deeper into inference and semis.
Introducing The Working Knowledge, a new technical publication from SeedToScale by Accel to help founders and technical teams make better decisions about the products they build. It brings together AI research and lessons from practitioners in playbooks teams can put to work. Our first, “How to Build a Vertical AI Product” by Tarun Raheja, explores how to evaluate models for your domain, build the context and tools around them, and make the product reliable in production. With a framework and worked example, it also helps teams decide what to keep, simplify or remove as models improve. Explore The Working Knowledge and download the playbook: seedtoscale.com/the-working-…
1
2
13
590
Tarun Raheja retweeted
A model can ace a demo on Monday and still fail a customer on Tuesday. If you’re building vertical AI, you know how much work lives between those two moments. Understanding the customer’s workflow. Giving the model the right context and tools. Catching the failures a polished demo never shows. And just when you’ve figured some of it out, a better model arrives. What do you keep? What do you rebuild? What can you finally remove? These are the questions behind the first playbook from The Working Knowledge, a new technical publication from @Seed2Scale by @AccelIndia. Written by @atemyipod, its first-ever edition offers a practical way to build the system around a model, test whether it helps, and make the result reliable enough for real customers. The kind of read we’d bring into a room of founders and ask: 𝗪𝗵𝗮𝘁 𝗵𝗮𝘃𝗲 𝘆𝗼𝘂 𝗹𝗲𝗮𝗿𝗻𝗲𝗱 𝘁𝗵𝗲 𝗵𝗮𝗿𝗱 𝘄𝗮𝘆? Uncover the playbook, then bring the question back to your team: seedtoscale.com/the-working-…
1
6
399
Tarun Raheja retweeted
Two teams can use the same frontier model and ship very different AI products. The difference is the harness around it. Prompts, context, tools, memory, permissions and checks decide whether an uneven model is reliable in production. The Working Knowledge is SeedToScale's editorial property on ground-level thinking for AI. Its debut working paper, authored by @atemyipod, breaks down how to engineer each part, and how to test every harness change before it ships. Read the full working paper → seedtoscale.com/the-working-…
1
2
7
479