ResearchVault

AI-powered research paper analyser. Upload PDFs, extract structured metadata automatically, chat with your paper library, and compare papers side by side.

Built with FastAPI, Next.js 14, MongoDB, ChromaDB, and the Claude API.

---

Features

• Auto-extraction — upload a PDF and Claude extracts all metadata: authors, architecture, datasets, metrics, results, limitations, and more
• Paper type detection — automatically classifies papers as empirical, survey, or theoretical and applies the right schema
• RAG chatbot — ask natural language questions across your entire library with cited, section-level answers
• Paper comparison — compare 2–4 papers (or all papers at once) across architecture, techniques, metrics, and results with a side-by-side table and bar chart
• Dataset tracking — datasets are extracted and stored separately, linked to the papers that use them
• Semantic search — search papers by description using embedding-based retrieval

---

Tech stack

Layer	Technology
Frontend	Next.js 14 (App Router), TypeScript, Tailwind CSS, Recharts
Backend	FastAPI, Python 3.10, motor (async MongoDB)
Database	MongoDB Atlas
Vector store	ChromaDB + sentence-transformers (all-MiniLM-L6-v2)
AI	Anthropic Claude API (claude-sonnet-4-6)
PDF parsing	PyMuPDF
Package manager	uv (backend), npm (frontend)

---

Project structure

ResearchVault/
├── Backend/
│   ├── app/
│   │   ├── models/          # Pydantic schemas
│   │   ├── routes/          # FastAPI routers (ask, compare, ingest, papers)
│   │   ├── services/        # claude_service.py — all Claude API logic
│   │   ├── database.py      # MongoDB motor connection
│   │   ├── embeddings.py    # ChromaDB + sentence-transformers
│   │   └── main.py          # FastAPI app entry point
│   ├── .env                 # API keys (never commit this)
│   └── pyproject.toml
└── frontend/
    ├── app/                 # Next.js App Router pages
    │   ├── page.tsx         # Library
    │   ├── upload/          # Upload page
    │   ├── papers/[id]/     # Paper detail
    │   ├── chat/            # RAG chatbot
    │   ├── compare/         # Paper comparison
    │   └── datasets/        # Dataset browser
    ├── components/          # Sidebar, PaperCard
    ├── lib/api.ts           # Typed fetch functions
    └── .env.local           # Frontend env (never commit this)

---

Local setup

Prerequisites

• Python 3.10+
• Node.js 18+
• uv (pip install uv)
• MongoDB Atlas account (free tier works)
• Anthropic API key

1. Clone the repo

git clone https://github.com/FriedIce-623/ResearchVault.git
cd ResearchVault

2. Backend setup

cd Backend
uv venv
uv sync

Create Backend/.env:

ANTHROPIC_API_KEY=sk-ant-your-key-here
MONGODB_URI=mongodb+srv://your-connection-string
DATABASE_NAME=researchvault

Run the backend:

uv run uvicorn app.main:app --reload --port 8000

Test it:

curl http://localhost:8000/health
# → {"status":"ok","papers":0,"datasets":0}

3. Frontend setup

cd frontend
npm install

Create frontend/.env.local:

NEXT_PUBLIC_API_URL=http://localhost:8000

Run the frontend:

npm run dev

Open http://localhost:3000

---

API endpoints

Method	Endpoint	Description
GET	/health	Backend health check
POST	/api/ingest/upload	Upload and extract a PDF
GET	/api/papers	List all papers
GET	/api/papers/search?q=	Semantic search
GET	/api/papers/{paper_id}	Get paper by ID
PATCH	/api/papers/{paper_id}	Update paper fields
DELETE	/api/papers/{paper_id}	Delete paper
GET	/api/papers/datasets/all	List all datasets
POST	/api/ask	Ask a question (RAG)
POST	/api/ask/stream	Streaming answer
POST	/api/compare	Compare papers

---

Paper schemas

Fields vary by paper type. All types share: paper_id, name, authors, doi, link, date_of_publication, code_link, key_insights, limitations, metrics_used, results.

Empirical adds: architecture, key_techniques, preprocessing, training_strategy, dataset_ids

Survey adds: papers_surveyed, taxonomy, research_gaps, time_period_covered

Theoretical adds: propositions, proofs_or_derivations, assumptions, applicability

Dataset schema: dataset_id, dataset_name, paper_ids, public, samples, support, classes, task, modality, link, key_insights

---

Environment variables

Backend (Backend/.env)

Variable	Description
ANTHROPIC_API_KEY	Your Anthropic API key
MONGODB_URI	MongoDB Atlas connection string
DATABASE_NAME	MongoDB database name (e.g. researchvault)

Frontend (frontend/.env.local)

Variable	Description
NEXT_PUBLIC_API_URL	Backend URL (e.g. http://localhost:8000 locally, your Render URL in production)

---

Deployment

Frontend is deployed on Vercel. Backend is deployed on Render.

See the repo for deployment configuration.

---

Notes

• First PDF upload takes ~30 seconds extra — sentence-transformers downloads its model (~90MB) once and caches it
• Compare all caps at 20 papers to stay within Claude's context window
• ChromaDB persists to Backend/chroma_db/ on disk — this folder is gitignored