Hello
Welcome to my blog! I share some interesing projects and thoughts here.
The Simulator Is the Reward: Building LLM Agents for Physical Engineering Domains
There is a class of software that millions of engineers use every day. It generates rich, deterministic, physically grounded feedback, and the LLM agent community has almost ign...
Decode-Free Information Extraction: Evidence Selection in a Single Forward Pass
Evidence selection — deciding which passages are relevant to a query — sits at the heart of RAG pipelines, open-domain QA, and multi-hop reasoning. In this project I ask a simpl...