About This Project
Overview
TechRadar Agent turns public research into an interrogable knowledge base: it ingests sources from arXiv, OSTI, FERC, NERC, and GitHub into a typed schema, chunks and embeds them into ChromaDB with bucket-scoped retrieval, and exposes kb_search, kb_get_document, and kb_status as MCP tools so Claude Code can query it directly. A multi-step research agent then composes evaluated, multi-source briefs on demand — every response stamped with a data-freshness timestamp.
Motivation
I wanted a way to actually keep up with energy-systems and AI-infrastructure research without spending hours a week manually skimming arXiv, FERC filings, and GitHub repos. The idea was to build something that could ingest, index, and reason over that firehose the same way I'd want a research assistant to.
Lessons Learned
Bucket-scoped retrieval turned out to matter more than raw embedding quality — without pre-filtering by source type, regulatory documents kept getting drowned out by the much larger volume of arXiv papers. Exposing the whole thing as MCP tools also taught me a lot about designing tool schemas an LLM can actually call correctly.