LLM Interpretability
Black-box interpretability for language models: benchmarking, an open-source library, and a novel probing method.
Case study in progress
Vaibhav
Agarwal
I build reliable AI systems — where LLM infrastructure meets real backend engineering.
Writing about RAG, agents, evaluation, and the infrastructure that makes AI reliable in production. Notes from building, not tutorials from docs.
- Now Building at Monaco (opens in a new tab) (AI CRM)
- Before Siftly (opens in a new tab) (YC)
- MS AI Georgia Tech
- BTech EE IIT Kanpur
01 / Work
All projects →Selected projects
MCP Servers
Model Context Protocol servers that expose structured data layers to LLM tools cleanly and reliably.
Case study in progress
AI News Desk
A personal aggregation pipeline that tracks the AI landscape and auto-drafts posts for distribution.
Case study in progress
02 / Writing
All writing →Recent writing
Current series · Production RAG
- Production RAG: the complete guidePillar
- Chunking strategies for RAG: fixed-size vs recursive vs semantic (benchmarked)Soon
- pgvector vs a dedicated vector database: when do you actually need one?Soon
- How to evaluate RAG retrieval quality (the metrics that matter)Soon
- Does reranking actually improve RAG answers? (I measured it)Soon
03 / Connect
Say hello
Working on retrieval, agents, or putting LLMs into production? I'm always up for comparing notes. Email is best: vaibhavkumaragarwal2020@gmail.com