Rerank with Conversation History
thisguyknowsai · x · 2026-07-11
This covers Technique 3: Context-Aware Reranking.
Common practice is to retrieve top-10 chunks and feed them all to the model. The improvement here uses a lightweight reranker that scores chunks based on full conversation history rather than just the current query. Content contradicting prior context is downweighted, while content continuing the context is upweighted.
The author claims this reduces contradictions in final output by 28%.
More from coding & agent
- He benchmarks new models by having them build their own copilot — $28 of compute — bucolucas · 2026-10-27
- Should agents be allowed to change plans mid-task? Reddit debates the drift boundary — wistfulcracker84 · 2026-10-27
- HeyGen + Canva MCP turns slide speaker notes into an avatar presentation video — HeyGen · 2026-10-27
- Local beats cloud: Qwen 3.8 FN handles private student data as well as frontier models — dr_lm · 2026-10-27
- Cloudflare Computer adds agent interruption recovery and approval pauses in v0.7.0 — irvinebroque · 2026-10-27
- LLM Tails cuts agent input tokens 80% by making agents read before they write — savadks1818 · 2026-10-27