Daksh Gupta

Daksh Gupta is co-founder and CEO of Greptile, which reviews pull requests with full-repository context instead of diff-only comments. He started the company with Georgia Tech classmates after graduation, entered Y Combinator, and pivoted from a codebase-context API into live AI code review when customers showed him where the pain really was.

Georgia Tech classmates and a SF Airbnb

In a Hatchline interview he said he met co-founder Soohoon Choi in a startup group-project class at Georgia Tech. Choi took it because he had read Paul Graham; Gupta took it "because it had the fewest hours needed to fulfill the capstone requirement." They enjoyed shipping bad ideas together, then convinced Gupta's roommate Vaishant Kameswaran to move to San Francisco the day after graduation. Angel money from Chris Klaus and others covered a rough Airbnb while they applied to YC.

The context-API pivot to PR review

The team entered Y Combinator's Winter 2024 batch. Early Greptile was a codebase context product meant to teach language models about large repositories so any coding tool could plug in. Customers used that API to build homemade AI reviewers. Talking to them, Gupta told Hatchline, convinced the founders that "Everyone would need to use AI to review their code, as we will produce far too much code to review as models improve," and that home-built reviewers were "harder to get right than they looked." Within a few months they pivoted the whole company to AI pull-request review with repository-wide context.

Full-repo review as the product

Greptile now sits in ordinary engineering workflows as an independent review layer beside human reviewers and coding agents. Gupta's public product case is that useful comments need full-repo understanding, team-specific standards, and a way to separate signal from nitpicks. On Hatchline he said the question he wishes more people asked is "How we actually separate signal from noise in AI code review," because useful feedback is a product and modeling problem, not a weekend script.

Why coding stays human and review can automate

His contrarian industry view, stated on Hatchline, is twofold: "AI will never fully automate coding because humans are not good enough at articulating exactly what they want upfront," so bugs remain in the human-AI loop; and "Catching bugs is a tractable problem; a piece of code either has a bug or it doesn't. It does not require opinion." That is why he believes validation can become autonomous even if generation cannot.