🚀 ResearchLens

ResearchLens is an AI-powered research assistant that allows users to interact with research papers using natural language.

Upload a paper or choose from default datasets, ask questions, and get context-aware answers powered by Retrieval-Augmented Generation (RAG).

---

🌐 Live Demo

👉 https://researchlenss.streamlit.app/

---

✨ Features

• 📄 Ask questions about research papers
• 📤 Upload your own PDF and query it instantly
• 📚 Switch between multiple research papers
• 🔍 Semantic search using FAISS
• 🧠 Cross-encoder reranking for better relevance
• 🤖 LLM-based answer generation
• 💬 Chat-style interface (like ChatGPT)
• ⚙️ Adjustable answer length

---

🧠 How It Works (RAG Pipeline)

PDF → Text Extraction → Chunking → Embeddings → FAISS Index  
→ Retrieval → Reranking → LLM → Answer

---

🏗️ Architecture

• Text Extraction: Extracts raw text from PDFs
• Chunking: Splits text into meaningful segments
• Embeddings: Converts text into vector representations
• FAISS Index: Enables fast similarity search
• Retriever: Finds relevant chunks for a query
• Reranker: Improves relevance using cross-encoder
• LLM: Generates final answer from retrieved context

---

🖥️ Demo Use Cases

Try asking:

• What is the main contribution of the paper?
• What architecture does the paper propose?
• What dataset was used?
• How does the model work?

---

⚙️ Tech Stack

• Frontend: Streamlit
• Backend: Python
• Vector DB: FAISS
• Embeddings: Sentence Transformers
• LLM: FLAN-T5 / TinyLlama (local)
• Reranking: Cross-encoder

---

🚀 How to Run Locally

1. Clone the repo

git clone https://github.com/your-username/researchlens.git
cd researchlens

2. Install dependencies

pip install -r requirements.txt

3. Build index (optional)

python -m src.indexing.build_index

4. Run the app

streamlit run app/streamlit_app.py

---

📂 Project Structure

app/
  └── streamlit_app.py

src/
  ├── data/
  ├── embeddings/
  ├── indexing/
  ├── models/
  ├── retrieval/

data/
  └── papers/

---

⚠️ Deployment Notes

• Lightweight model used for cloud deployment
• Larger models (TinyLlama) used locally for better performance
• Handles CPU-only environments

---

🔮 Future Improvements

• Inline citations in answers
• Hybrid search (BM25 + embeddings)
• Multi-document memory
• Improved UI/UX
• Better LLM integration

---

💡 Key Learnings

• Built a full RAG pipeline from scratch
• Solved real-world deployment issues (CPU, memory limits)
• Designed scalable multi-document architecture
• Balanced model performance vs resource constraints

---

👨‍💻 Author

Ronak Das

---

⭐ If you like this project

Give it a star on GitHub ⭐