Dev builds research agent with mandatory citations and an LLM judge that strips weak ones
FreakFrakFrok · reddit · 2026-10-07
A developer built a personal research agent that answers only from user-picked sources (news sites, consultancies, company sites, YouTube), with every claim cited. A second LLM pass acts as a judge, checking each citation against retrieved text and stripping ones that don't hold up.
Notable design choices:
- Queue-based LLM work: the web app only enqueues tasks; workers run a local model with SKIP LOCKED and replicas—cheap and private, but no token streaming, just polling.
- Memory via "spaces": notes, videos and docs are saved to spaces and retrieved alongside your feed.
The author is crowdsourcing adversarial testing—bogus citations, missing "I don't know", UX friction—and offers 15,000 extra credits for harsh feedback.
More from coding & agent
- Dev open-sources AutoType, a free Wispr Flow alternative for Linux voice typing — premakin · 2026-10-07
- Running two Muse AI agents as a boss-worker digital production factory — NickPassig · 2026-10-07
- Coinbase's Lincoln Murr on paying AI agents via the x402 protocol — MurrLincoln · 2026-10-07
- Resend logs 3M MCP calls last month, up ~30x since April — dsp_ · 2026-10-07
- Dev vibes-clones his 2-year game with Claude: looks right, plays wrong — IanArawjo · 2026-10-07
- Y Combinator-backed Linc launches discovery layer that turns enterprise tribal knowledge into SOPs — ycombinator · 2026-10-07