Project · 2026
Soma Agent
Agentic document search using a ReAct loop that reasons about what to read instead of relying on vector similarity. Small embedding model (90MB), agent reasoning compensates. 50% latency reduction over naive RAG.
Overview
Agentic document search using a ReAct loop that reasons about what to read instead of relying on vector similarity. Small embedding model (90MB), agent reasoning compensates. 50% latency reduction over naive RAG.
Problem
A semantic search result alone may not identify which documents actually answer a user’s question.
Approach
Built a ReAct-style document-search agent that decides which sources to search and read, using a lightweight embedding model and tool use.
Results
The project reports a 50% latency reduction against its naive RAG comparison.