Add HF Spaces metadata and project README

README.md

@@ -1,2 +1,59 @@
-# project-recon
-Multi-Agent Research Navigator - LangGraph + Semantic Scholar
+---
+title: RECON
+emoji: 🔍
+colorFrom: blue
+colorTo: indigo
+sdk: gradio
+sdk_version: "5.0.0"
+app_file: app.py
+pinned: true
+license: mit
+short_description: Multi-agent ML literature research with staleness detection
+---
+
+# RECON — Multi-Agent Research Navigator
+
+**Temporally-aware ML literature research. Live Semantic Scholar. Staleness detection.**
+
+RECON is a four-agent LangGraph system that retrieves live ML papers, evaluates evidence quality using a four-verdict critic, and synthesizes research positions with per-claim confidence scoring.
+
+## What makes it different from standard RAG
+
+Standard RAG retrieves the most semantically similar chunk with no mechanism to detect whether that chunk has been superseded. A 2019 paper cited 600 times and never contradicted is strong evidence. A 2019 paper that a 2023 paper explicitly refutes is weak evidence — regardless of its cosine similarity score. RECON's critic reasons about this distinction.
+
+## Architecture
+
+```
+session_loader → planner → retriever → critic → synthesizer
+                                           ↓ STALE/CONTRADICTED/INSUFFICIENT
+                                     retry_retriever → critic (max 2x)
+```
+
+**Four agents:**
+- **Planner** — decomposes query into temporally-typed sub-questions (foundational / recent / contested)
+- **Retriever** — searches Semantic Scholar (200M+ papers) + DuckDuckGo with hybrid scoring
+- **Critic** — four-verdict taxonomy: PASS / STALE / CONTRADICTED / INSUFFICIENT
+- **Synthesizer** — structured position with inline citations and per-claim confidence
+
+## Eval results (130-question ground truth dataset)
+
+| Architecture | Position Acc | Staleness Catch | Latency |
+|---|---|---|---|
+| Single-agent RAG | 32.3% | 0% | 4.8s |
+| Naive multi-agent | 44.6% | 0% | 23.9s |
+| **RECON (linear decay)** | **43.9%** | **52%** | **17.1s** |
+
+RECON catches 52% of superseded claims vs 0% for single-pass RAG on Category B questions sourced from real survey paper supersession chains.
+
+## Tech stack
+
+- **Orchestration:** LangGraph
+- **LLM:** Groq / LLaMA 3.3-70B
+- **Retrieval:** Semantic Scholar REST API + DuckDuckGo
+- **Embeddings:** all-MiniLM-L6-v2
+- **Session memory:** SQLite
+- **Eval:** Ragas + LLM-as-judge
+
+## GitHub
+
+[github.com/MukulRay1603/project-recon](https://github.com/MukulRay1603/project-recon)