Daksh Gupta

Gupta met his cofounders at Georgia Tech. Their CREATE-X project began as an AI shopping assistant called Tabnam, then shifted toward helping developers understand software. He describes wanting a tool that could learn a codebase and help with debugging and quality assurance. Greptile emerged from that work and continued through Y Combinator’s winter 2024 program.[2][4]

Greptile’s current product reviews pull requests using context from the repository and connected development tools. Its website describes investigating the affected code and testing proposed changes rather than treating a diff as an isolated text passage. Gupta leads the company with his cofounders and team; the review system is their product.[1][4][6]

Measure generated code using several signals

In his AI Engineer talk, Gupta compares agent-associated and human-written pull requests using revert rates, issue severity and review rounds. He explains how his analysis identifies generated changes through metadata and other signals. These observations concern the reviewed repositories and detection method, rather than proving that all generated code has the same quality as all human code.[5]

Ask what a change does to the application’s contract

The talk connects review to three questions: whether a change breaks expected behavior, makes a future break more likely or fulfills its author’s intent. Gupta uses blast-radius analysis and sandboxed browser testing to illustrate the approach. This makes code validation a question about application behavior and surrounding dependencies, extending beyond whether a patch compiles.[5]