Building a RAG-powered FAQ bot: RRF-fused dedup keeps the knowledge base clean
Al_Grigor · x · 2026-10-06
A developer serving thousands of students in DataTalksClub's free courses shared his RAG-powered FAQ system:
- Answer flow: retrieves similar questions from an existing FAQ index, uses an LLM to generate answers, and replies directly in Slack.
- Dataset dedup: every new FAQ proposal triggers 2 searches — one on the question, one on the full FAQ record — fused with RRF, then an LLM classifies it as NEW / UPDATE / DUPLICATE / WRONGCOURSE. Only NEW and UPDATE reach a human reviewer.
- Tracing: uses Opik by Comet to trace and evaluate the workflow.
A practical blueprint for anyone maintaining an LLM-assisted FAQ or knowledge base at scale.
More from coding & agent
- OpenAI streamlines ChatGPT plugin submissions: upload zip, fix validation, publish — Dimillian · 2026-10-06
- Long-running benchmarks find Strata inference server failing full-build scenarios — julianharris · 2026-10-06
- OpenAI makes Codex Auto-review free for all: a second agent vets high-risk actions — kimmonismus · 2026-10-06
- SourceLearn Builds Source-Specific Agent Competence, Wins 13 of 15 Benchmarks — GeorgiaTech · 2026-10-06
- Programmatic Search Agents boost task success by up to 7.56 points over query-based agents — _reachsumit · 2026-10-06
- SWE-Race: 188 real concurrency bugs benchmark where GPT-5.6 Luna scores 81% — heyitsdannyle · 2026-10-06