🔬 PRISM

Paper Retrieval and Intelligent Summarization Model

PRISM (Paper Retrieval and Intelligent Summarization Model) is an AI-powered Research Paper Assistant built using Retrieval-Augmented Generation (RAG) architecture. It enables researchers, students, reviewers, and academicians to upload one or multiple research papers (PDFs), retrieve semantically relevant information, and generate intelligent, citation-aware research-focused answers.

Instead of manually reading hundreds of pages, PRISM helps users interact with papers conversationally and extract valuable insights instantly.

---

📌 Why PRISM?

Reading research papers is time-consuming.

Researchers often spend hours trying to:

• Understand the methodology
• Identify limitations
• Find research gaps
• Compare multiple papers
• Write review comments
• Extract contributions
• Analyze novelty

PRISM solves this problem by combining semantic retrieval with large language models.

It makes research paper understanding:

✅ Faster ✅ More structured ✅ More explainable ✅ More interactive ✅ More accurate

---

🚀 What Problem Does It Solve?

Traditional PDF readers only allow reading.

PRISM transforms static PDFs into an intelligent searchable knowledge base.

It solves:

1. Information Overload

Long papers become queryable.

2. Time Consumption

Reduces reading time significantly.

3. Reviewer Assistance

Helps generate academic review comments.

4. Literature Review Support

Helps compare multiple papers.

5. Research Gap Identification

Detects unsolved areas in previous work.

---

⚙️ How PRISM Works

PRISM follows the RAG Pipeline:

PDF Upload
   ↓
Text Extraction
   ↓
Chunking
   ↓
Embedding Generation
   ↓
FAISS Vector Storage
   ↓
Semantic Retrieval
   ↓
Groq LLM Response Generation
   ↓
Citation-based Output

---

🧠 Core Architecture

1. PDF Processing

Research papers are uploaded in PDF format.

PRISM extracts text using:

• PyPDF

This allows PRISM to read each page individually.

---

2. Intelligent Chunking

Large text is divided into smaller chunks.

Used:

• RecursiveCharacterTextSplitter

Why?

Because LLMs work better on smaller contextual segments.

---

3. Semantic Embeddings

Each chunk is converted into vector embeddings.

Used:

• Sentence Transformers
• all-mpnet-base-v2

Why?

This captures semantic meaning instead of keyword matching.

---

4. FAISS Vector Database

Embeddings are stored inside FAISS.

Why?

Because FAISS enables ultra-fast similarity search.

This allows retrieval of the most relevant chunks.

---

5. Groq LLM Integration

After retrieval:

Relevant chunks → sent to Groq LLM.

Groq generates:

• Accurate answers
• Research summaries
• Reviewer comments
• Comparative analysis

---

✨ Features

📄 Multi-PDF Support

Upload multiple research papers simultaneously.

Useful for:

• Literature surveys
• Comparative studies

---

🔍 Semantic Search

PRISM retrieves relevant content using meaning-based similarity.

Not keyword-based.

More intelligent retrieval.

---

📚 Citation-based Responses

Every important answer includes:

• File name
• Page number
• Chunk number

This improves transparency.

Example:

(Source: SPASHTA_CVIP_2025.pdf, Page: 22, Chunk: 35)

---

🧪 Research-focused Action Modes

PRISM supports:

Summarize Paper

Complete paper summary.

Extract Methodology

Find technical methods used.

Extract Limitations

Identify weaknesses.

Extract Future Scope

Find improvement directions.

Extract Research Gap

Identify unsolved problems.

Estimate Novelty

Estimate uniqueness of contribution.

Extract Contributions

Identify key contributions.

Generate Related Work Summary

Understand literature position.

Generate Reviewer Comments

Simulate academic reviewer feedback.

Compare Multiple Papers

Compare papers structurally.

---

💬 Session Chat Memory

Maintains chat history for contextual conversation.

Enables follow-up questions.

---

📥 Download Responses

Generated answers can be downloaded for later use.

---

🛠 Tech Stack

Technology	Purpose
Python	Core development
Streamlit	Frontend/UI
Groq API	LLM inference
FAISS	Vector database
Sentence Transformers	Embeddings
PyPDF	PDF text extraction
LangChain Splitters	Chunking
NumPy	Vector operations

---

📂 Project Structure

PRISM/
│── app.py
│── utils.py
│── requirements.txt
│── README.md
│── .gitignore
│── screenshots/
│   ├── home.png
│   ├── upload.png
│   ├── response.png
│   ├── source_chunks.png
│── .streamlit/
│   └── secrets.toml

---

📸 Screenshots

Home Interface

---

Upload & Processing

---

AI Generated Response

---

Source Chunk Retrieval

---

🔐 Setup Instructions

Clone Repository

git clone https://github.com/build-with-saurav/PRISM.git
cd PRISM

---

Create Virtual Environment

python3 -m venv prism_env
source prism_env/bin/activate

---

Install Dependencies

pip install -r requirements.txt

---

Add API Key

Create:

.streamlit/secrets.toml

Add:

GROQ_API_KEY="your_api_key_here"

---

▶️ Run Project

streamlit run app.py

---

📈 Future Improvements

• OCR support for scanned PDFs
• Persistent FAISS storage
• Research paper recommendation engine
• Citation graph generation
• Auto literature review generation
• Research plagiarism similarity checker
• Export to PDF
• Multi-user collaboration

---

💡 Why This Project Is Valuable

PRISM demonstrates:

• Retrieval-Augmented Generation (RAG)
• Vector Databases
• Semantic Search
• LLM Prompt Engineering
• Document Intelligence
• Applied NLP
• Research Automation

This makes it highly valuable for:

• AI Engineers
• ML Engineers
• GenAI Engineers
• Research Engineers
• NLP Engineers

---

👨‍💻 Author

Saurav Kumar Singh

B.Tech CSE — NIT Calicut AI/ML Engineer | Research Enthusiast | GenAI Builder

---

Star ⭐ the repository Fork 🍴 the project Contribute 🚀 improvements