Semantic Search & RAG for the Scientific Community.
Navigating the arXiv repository can be overwhelming. Traditional keyword search often misses context, making it difficult for researchers to find papers based on concepts rather than just titles.
PaperMind is an intelligent research assistant that leverages Vector Search and Retrieval-Augmented Generation (RAG). It allows users to query scientific databases using natural language, retrieving contextually relevant papers and summarizing findings instantly.
Unlike standard regex matching, PaperMind uses embedding models to understand the intent behind your query.
- Natural Language Queries: Ask "How does attention mechanism work?" and get papers on Transformers.
- Vector Search: Powered by MongoDB Atlas Vector Search for high-dimensional similarity matching.
- Context-Aware Q&A: The application retrieves relevant paper abstracts and feeds them into an LLM context window to answer specific questions with citations.
- Blazing Fast UI: Built on Next.js 16 (App Router) for server-side streaming and instant interactions.
- Live arXiv Integration: Real-time access to the latest pre-prints.
- Self-Improving Results: User interactions help refine the vector search relevance over time.
This project is built with performance and scalability in mind, utilizing a modern Serverless architecture.
| Component | Technology | Reasoning |
|---|---|---|
| Framework | Next.js 16 | App Router for React Server Components & SEO. |
| Database | MongoDB Atlas | Unified document store + Vector Search engine. |
| Embeddings | HuggingFace | sentence-transformers/all-MiniLM-L6-v2 for efficient local embedding. |
| LLM Orchestration | Vercel AI SDK | Streamlined prompt engineering and streaming responses. |
| Inference | Groq | Ultra-low latency inference for the chat interface. |
| Styling | Tailwind CSS | Utility-first styling for a responsive, modern UI. |
To run PaperMind locally, follow these steps:
git clone [https://github.com/alangeirnaert/papermind.git](https://github.com/alangeirnaert/papermind.git)
cd papermindnpm install
# or
pnpm installCreate a .env.local file and add your keys:
MONGODB_URI=...
GROQ_API_KEY=...
HF_TOKEN=...npm run devOpen http://localhost:3000 to view it in the browser.
Current focus: Performance Optimization & User Personalization.
- 🧠 AI Summarizer – One-click "TL;DR" for lengthy PDFs.
- ⚡ Backend Optimization – Caching frequent arXiv queries to reduce API latency.
- 🌍 Custom Domain – Enhanced branding and SEO.
- Data Source: arXiv API
- Search Engine: MongoDB Atlas Vector Search
- Model: Hugging Face Transformers
Alan Geirnaert Full-Stack Engineer & Physics Enthusiast
This project is open-source under the MIT License. Use it, fork it, launch your own scientific tools.
If you find this project useful, please give it a star ⭐ on GitHub!