PRISM — Personalized Retrieval-based Intelligent Study Mentor

PRISM is an AI-powered learning system that helps users understand knowledge from books and long-form content without reading everything themselves.

The system retrieves relevant information from stored knowledge and generates personalized explanations based on the user’s learning style and level.

This project is built as a real system, not a demo, and demonstrates retrieval, memory, and reasoning using Qdrant and a modern LLM.

---

🚩 Problem Statement

Many people want to learn from books but struggle with:

• Lack of reading habits
• Time constraints
• Different learning preferences

Books contain valuable knowledge, but the format is not always accessible to everyone.

PRISM addresses this by:

• Extracting knowledge from books
• Storing it in a searchable memory
• Explaining concepts in a personalized way using AI

---

💡 Solution Overview

PRISM uses a Retrieval-Augmented Generation (RAG) approach:

1. Book content is stored as semantic embeddings
2. Relevant knowledge is retrieved based on the user query
3. An AI agent builds a personalized explanation prompt
4. A language model generates a human-friendly explanation

This ensures answers are grounded in real content, not hallucinated.

---

🧠 System Architecture

The system is divided into the following layers:

• Data Ingestion: Load, chunk, and embed book text
• Vector Memory: Store embeddings in Qdrant
• Retrieval & Agent Reasoning: Fetch relevant context and personalize explanations
• LLM Generation: Generate natural language explanations

Detailed architecture is documented in ARCHITECTURE.md.

---

🧩 Key Technologies

• Python
• Qdrant — Vector database for semantic search and memory
• SentenceTransformers — Text embeddings
• Gemini 2.5 Flash — Language model (free tier)
• Docker — Running Qdrant locally

---

🧠 Why Qdrant is Critical

Qdrant is the backbone of PRISM’s memory system.

It enables:

• Semantic search over book knowledge
• Fast retrieval of relevant concepts
• Separation of memory from reasoning
• Scalable, production-like architecture

Without Qdrant, the system would degrade into a simple prompt-based chatbot.

---

📂 Project Structure

prism/ │── src/ │ ├── load_book.py │ ├── chunk_book.py │ ├── embed_chunks.py │ ├── store_in_qdrant.py │ ├── search_qdrant.py │ ├── agent_explain.py │ └── llm_runner.py │ │── data/ │ └── books/ │ └── atomic_habits.txt │ │── ARCHITECTURE.md │── README.md │── requirements.txt │── .gitignore

---

⚙️ Setup Instructions

1️⃣ Clone the repository

git clone <repository-url>
cd prism

2️⃣ Create and activate virtual environment

python -m venv venv
venv\Scripts\activate

3️⃣ Install dependencies

pip install -r requirements.txt

4️⃣ Start Qdrant (Docker)

docker run -d -p 6333:6333 qdrant/qdrant

Open dashboard:

http://localhost:6333/dashboard

🔐 Environment Variables Create a .env file in the project root:

GEMINI_API_KEY=your_api_key_here

Ensure .env is added to .gitignore

▶️ Running the System Load data into Qdrant

python src/store_in_qdrant.py

Run the AI Tutor

python src/llm_runner.py

🧪 Example Query

How do habits become part of identity?

The system retrieves relevant book content and explains it in a story-based style for beginners.

⚠️ Limitations & Ethics

Limitations

• Depends on quality of source text
• Free-tier LLM availability may vary

Ethics

• No personal user data stored
• Transparent retrieval-based explanations
• Reduced hallucinations via grounding

🚀 Future Enhancements

• Audio-based explanations
• Adaptive difficulty levels
• User learning progress memory
• Multi-book support
• Web or mobile interface

📌 Summary

• PRISM demonstrates a complete AI system combining:
• Vector memory
• Semantic search
• Agent-based reasoning
• LLM-powered explanations
• It is designed to be extensible, explainable, and aligned with real-world AI system principles.