Abhinav Sinha

Abhinav Sinha is cofounder and CEO of Lucidic AI, a San Francisco company that helps teams observe, debug, test, and evaluate AI agents in production. Lucidic visualizes an agent's full journey - API calls, decision nodes, and failure modes - and adds session-level rubrics so teams measure agent runs against their own criteria. The company joined the YC Winter 2025 batch.

Stanford friends who started with Smash Bros

Sinha studied computer science with an AI specialization at Stanford for both his bachelor's and master's degrees and did research at the Stanford AI Lab. He worked as a software engineer at Apple and as a quant at Citadel Securities and Susquehanna. Cofounders Andy Liang and Jeremy Tian also came through Stanford CS with AI focus; YC's launch note says the three met playing Super Smash Bros as freshmen and later worked on deep learning research together.

Agents need session-level evals

On Launch HN, the Lucidic team argued that classic LLM-ops evals score prompts in isolation, while agent failures often come from memory, tools, or earlier steps in a long session. Their answer is custom rubrics with weighted criteria, plus an investigator agent that reviews those criteria against full runs. They also talk about "time travel" captures of agent state so developers can change context or tools and replay from that moment.