Esha's Twitter Thread

We have raised $61M, but that’s not what excites me. Our original idea was fine tuning LLMs for enterprises. Although we topped some benchmarks, the business was not viable.

And then we ended up in customer support – not because we were passionate about it, but because we just really wanted to change the world with AI. Customer support is one of the few places where you can make a lot of impact using AI right now. Our vision is building an

We are hiring. If any of the problems below excite you, apply at http://giga.ai/careers - How do we build an AI system which hallucinates less than 0.01% of the time? ChatGPT currently hallucinates 4-5% of the time (read more here) - How do we scale our infra to 100s of

P.S. $10k for any engineering referral we hire. Tag the best engineer you know

@eshamanideep Fat one looks like Mario’s evil uncle wario but the Indian version

@eshamanideep What the fuck are you saying if you didn’t have the CC I wouldn’t understand. That’s the truth not being a dick but fuck me.

@eshamanideep Curious why the fine-tuning biz was not viable? Not enough enterprises that cared to own a model? Shitty margins? Something else?

@eshamanideep Fired up today! Just getting started

@eshamanideep @eshamanideep is one of the brightest minds in AI but only if he gets enough sleep :)

@eshamanideep What a pleasure it has been working with you Esha! I'm excited to see how far we can push the product

@eshamanideep Enterprise fine-tuning faces a fundamental misalignment: the value enterprises seek (domain-specific accuracy, control, compliance) requires proprietary data at scale, but data quality and labeling costs kill unit economics. Most realize too late that RAG + prompt engineering

@eshamanideep Love the honesty. Chasing benchmarks is easy — building something viable is the real game. Excited to see how you pivot from performance to purpose. 🚀

@eshamanideep Damnnnnn

@eshamanideep The sub-0.01% hallucination requirement fundamentally changes the architecture. You can’t rely on probabilistic generation alone – need constrained decoding + retrieval + confidence scoring in the hot path. At 100M+ calls/day with 400ms latency budget, that means: •Inference

@eshamanideep Still taking angels?

@eshamanideep This video with no volume creeped me out

@eshamanideep Congrats people. We may very well sign up

@eshamanideep Not visible bc employees don’t want AI

@eshamanideep honesty matters 😎

@eshamanideep Bruh you looking for distribution ... media and marketing? Happy to build and execute the full strategy. Will make the GIGA name recognizable by everyone on X! Your video was fire 🔥🔥 the guys did it ... deserves a raise.

@eshamanideep incredible stuff, love this !

@eshamanideep CONGRATS BROS!

@eshamanideep 🔥🔥🔥

@eshamanideep Congratulations 👏👏🤩🤩

@eshamanideep ✨

@eshamanideep OG play. Real builder move.

@eshamanideep this is a really top tier demo! congrats on the raise!

@eshamanideep Pretty cool! @garrytan is also pretty bullish!

@eshamanideep Congratulations

@eshamanideep Congratulations Esha. 🥳 more power to you guys. Are you guys hiring in India?

@eshamanideep Congrats!

@eshamanideep Good for you guys