Deepayan Chakraborty

Novonus trains robots for high precision manufacturing. this is how we do it. novonus.com

@ShreenabhA cool stuff, really impressive. just DM'd you!

Robot in SF, remote operator in India, 150 ms latency

visit novonus.com for details

Novonus is hiring a Founding Engineer. Expertise in Robot Learning (imitation learning, diffusion policy), sim-to-real (Isaac Sim/MuJoCo), hands-on experience deploying learned policies on robot arms, Physical AI Training Pipelines. Bonus: worked with EMG sensors, edge deployment. Completely remote. Higher than competitive equity 👀 Like and comment below. We'll reach out. DMs are open.

@cchevyyc @ruslanjabari yo

comment if you’re a solo founder like me!

@adrisaagarwal this is really interesting thanks for sharing

@deepayancx ai-2040.com

working out of the Softmax’s office today, and they’re doing something pretty different from most frontier AI labs. their core thesis is: the next bottleneck in AI isn’t raw intelligence. it’s social intelligence. instead of asking “can an agent solve a coding problem?”, they’re asking: - can multiple agents cooperate? - can they negotiate? - can they divide labor? - can they build trust? - can they handle conflicting goals? - can they form stable societies? they call this organic alignment: the idea that alignment emerges naturally when agents learn to cooperate, similar to how people form societies. to study this, they’ve built an open-source benchmark called Cogames, which is basically multiplayer games designed specifically for AI agents. think of it as: AlphaGo → one agent playing Go. Softmax → dozens of agents playing an MMO where cooperation matters more than individual intelligence. in their first game, agents have specialized roles like miners, scouts, and defenders. they need to coordinate, share resources, and compete against other teams. researchers can then measure things like: - leadership emergence, - specialization, - communication, - cooperation, - and strategic planning. the bet they’re making is that the future of AI isn’t one giant super-assistant. it’s thousands (or millions) of autonomous agents working together. if that’s true, today’s benchmarks (MMLU, SWE-Bench, coding evals, etc.) miss a huge part of what matters. we don’t really know how to evaluate whether agents can function as members of an organization. most labs focus on making individual models smarter; Softmax is trying to understand how intelligent systems behave collectively. if you’re interested in agents, multi-agent systems, or AI alignment, they’re one of the more interesting labs to watch right now

@NatalieLeviii Thanks Natalie!

@deepayancx So intriguing, great video!

Novonus trains robots for high precision manufacturing. this is how we do it. novonus.com

@daleverett thanks!

@deepayancx congrats on the launch gang

@DixitFlora6893 Thank you Flora

@deepayancx Very fascinating concept

@nisargxj @NatalieLeviii @obotiq Looks awesome 🙌

We built a robot to support caregivers, reduce burnout, and enhance care

holy shit the future of robotics genuinely is now. we now have actual robots moving like humans plucking grapes and screwing lightbulbs. we need more human-like robots in the world, especially in manufacturing where labor’s hard to find. to make robots like humans, you gotta capture the data that make us humans. that’s where we step in. novonus.com check it out 😉

sf has taught me everyday’s a new race to get as much done as possible. you start from square 1 and try and do as much as possible. some days feel less productive even though you’ve worked more, like when you spend a whole day sending out emails and recording videos rather than building the actual product. progress is progress as long as you’re getting closer to your goal, no matter how that may be.

@billgnofficial building a training pipeline for robots in high precision manufacturing: novonus.com would love to chat

I'm looking to connect with more founders who are building great and innovative products. If you're creating something exciting, let's connect, exchange ideas, and support each other's journey.

im trying to post smth everyday, build in public, and not have claude cowork do it for me. just me and my brain. brains are interesting, they provide every single useful signal for you to function. imo physical ai has to capture all of them to progress, and they’ve captured most of them. im capturing the one mostly overlooked: impedance. and using EMG sensors to do so, including all other vision, force and IMU sensors, to build the sensor rig for the most complete dataset and processing/deployment pipeline for robots to train on. it always comes back to the brain. also, check out my website: novonus.com

@DixitFlora6893 my agents finally work smart on the internet now 🙏

Your AI agent isn't dumb. It's just trying to read a website built for the human eye.

@cchevyyc i agree!

@deepayancx we have 5 more weeks but i already feel like we're running out of time. things move so fast it's both exciting and scary!

one thing I’ve realised building for a week in SF: it’s always back to basics. some of engineering’s greatest feats came from mimicking nature: sonar from bats, neural nets from the brain. nature, through billions of years of evolution, has performed experiments and perfected its design. time we start listening to it. physical AI’s next unlock is the same move. humans solve contact-rich manipulation effortlessly, and the signal is right there in the muscle. that’s what i’m targeting with Novonus: an imitation learning pipeline that teaches robots the delicate, high-precision assembly humans do without thinking using muscle intuition. here’s to 5 more weeks in SF, and to the basics.