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A multi-agent RAG system that detects when retrieved scientific evidence has been superseded by newer work.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Each paper in the results shows a reliability label: FOUNDATIONAL, CURRENT, DECLINING, or SUPERSEDED.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)