File size: 1,090 Bytes
3788f27 f4f6dd6 3788f27 4f24399 ca3ca45 3788f27 ca3ca45 3788f27 ca3ca45 3788f27 ca3ca45 3788f27 ca3ca45 0bf564b ca3ca45 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 | ---
title: RECON
emoji: 🔍
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.10.0
app_file: app.py
pinned: true
license: mit
short_description: Multi-agent ML literature research with staleness detection
---
> For full documentation, architecture details, and eval results: [GITHUB_README.md](./GITHUB_README.md)
# RECON — Temporally-Aware Scientific Retrieval
A multi-agent RAG system that detects when retrieved scientific evidence has been superseded by newer work.
**Try it:** Enter any research question. RECON retrieves papers from Semantic Scholar and OpenAlex, scores their reliability using a three-signal formula (citation centrality + recency + content coherence), and flags stale or contradicted evidence before synthesizing an answer.
Each paper in the results shows a reliability label: FOUNDATIONAL, CURRENT, DECLINING, or SUPERSEDED.
**Evaluation:** 44% staleness catch rate on a 130-question benchmark of real scientific supersession chains. Single-pass RAG baseline: 0%.
---
Built by Mukul Ray | [GitHub](https://github.com/MukulRay1603/project-recon)
|