Add YS0meone/Corvus — multi-agent academic research system

agents/YS0meone__corvus/README.md

@@ -0,0 +1,69 @@
+# Corvus — Multi-Agent AI Research System
+
+Corvus is a production-quality **multi-agent research assistant** built on LangGraph. It helps you discover academic papers and get evidence-based answers grounded in their full text.
+
+## What it does
+
+**1. Paper Finding**
+- Searches Semantic Scholar (200M+ papers) using keyword queries with filters (year, venue, citation count)
+- Runs web searches via Tavily to discover seminal work and broader context
+- Performs forward and backward citation snowballing to surface related papers
+- Auto-reruns searches with a refined plan if the initial results are insufficient
+
+**2. Evidence-Based Q&A**
+- Lets you select papers from the search results to focus on
+- Retrieves relevant evidence chunks from indexed full text (stored in Qdrant)
+- Synthesises grounded answers citing only the papers you chose
+- Clearly flags papers whose full text could not be indexed
+
+## Architecture
+
+```
+User → Supervisor (LangGraph)
+         ├─ Paper Finding Agent
+         │    ├─ Planner
+         │    ├─ ReAct search executor (s2_search, tavily, forward/backward snowball)
+         │    └─ Replanner
+         └─ Q&A Agent
+              ├─ Evidence retrieval (Qdrant vector search)
+              ├─ Sufficiency evaluator
+              └─ Answer synthesiser
+```
+
+- **Orchestration:** LangGraph supervisor + subgraph pattern
+- **LLM:** Google Gemini 2.0 Flash via LangChain
+- **Vector store:** Qdrant (scoped per selected papers)
+- **Ingestion:** Celery + Redis background worker — arXiv PDF → Grobid → chunked embeddings
+- **API:** FastAPI + LangGraph SDK streaming
+- **Frontend:** React 19 + Vite
+- **MCP server:** expose search tools directly in Claude Desktop
+
+## Key capabilities
+
+| Tool | Description |
+|------|-------------|
+| `s2_search_papers` | Semantic Scholar keyword search with filters |
+| `tavily_research_overview` | Web search for context and author profiles |
+| `forward_snowball` | Papers that cite a given seed paper |
+| `backward_snowball` | Papers cited by a given seed paper |
+| `retrieve_and_answer_question` | Evidence RAG over selected papers |
+
+## Live demo
+
+[corvus-agent.vercel.app](https://corvus-agent.vercel.app)
+
+## Quick start
+
+```bash
+git clone https://github.com/YS0meone/Corvus.git && cd Corvus
+cp backend/.env_example backend/.env.local
+# fill in OPENAI_API_KEY, GEMINI_API_KEY, COHERE_API_KEY, S2_API_KEY
+make infra        # starts Redis, Qdrant, Grobid via Docker Compose
+make worker       # Celery PDF ingestion worker
+make dev          # LangGraph backend  (http://localhost:2024)
+make frontend     # React frontend     (http://localhost:5173)
+```
+
+## License
+
+MIT — Copyright (c) 2025 Tim Yuan

agents/YS0meone__corvus/metadata.json

@@ -0,0 +1,14 @@
+{
+  "name": "corvus",
+  "author": "YS0meone",
+  "description": "Multi-agent AI research system — finds academic papers via semantic search & citation snowballing, then answers questions over them using agentic RAG.",
+  "repository": "https://github.com/YS0meone/Corvus",
+  "version": "1.0.0",
+  "category": "research",
+  "tags": ["academic-research", "multi-agent", "langgraph", "rag", "citation-snowballing", "semantic-scholar", "paper-search", "qa"],
+  "license": "MIT",
+  "model": "google:gemini-2.0-flash",
+  "adapters": ["system-prompt"],
+  "icon": false,
+  "banner": false
+}