DocMind

This repository contains an AI-powered learning assistant that allows users to upload PDF documents and interact with them using modern AI techniques such as semantic search, chat, summaries, and flashcards.

The project is intentionally structured to preserve multiple architectural versions of the same system, demonstrating engineering evolution, design trade-offs, and real-world deployment considerations.

---

📂 Repository Structure

AI-Learning-Assistant/
├── gemini_version/
│   ├── app.py                # Flask application using Google Gemini API
│   ├── requirements.txt      # Dependencies for Gemini version
│   └── README.md             # Setup instructions for Gemini version
├── groq_version/
│   ├── app.py                # Flask application using Groq API
│   ├── requirements.txt      # Dependencies for Groq version
│   ├── templates/            # Frontend HTML files
│   │   ├── index.html        # Landing page
│   │   ├── chat.html         # AI Chat interface
│   │   ├── summary.html      # Content summarization view
│   │   └── flashcards.html   # Generated study aids
│   └── README.md             # Setup instructions for Groq version
├── .gitignore                # Files to ignore in Git (e.g., .env, venv)
└── README.md                 # Project overview and documentation

---

📁 Folder Explanations

🔹 gemini_version/

This folder contains Version 1 of the project.

Purpose:

• Demonstrates a cloud-native AI architecture
• Uses Google Gemini and Google Cloud services
• Represents the initial implementation of the idea

Why it exists:

• Preserved intentionally to show:
  • Original design choices
  • Cloud-based AI integration
  • Comparison against later optimizations

Refer to gemini_version/README.md for full details.

---

🔹 groq_version/

This folder contains Version 2 of the project.

Purpose:

• A complete re-architecture focused on:
  • Cost efficiency
  • Free-tier deployment
  • Reduced cloud lock-in
• Designed to be production-deployable on Render

Key highlights:

• LLM inference using Groq (LLaMA 3.1)
• Local embeddings using SentenceTransformers
• Vector search using Supabase (PostgreSQL + pgvector)
• Full RAG-based chat, summaries, and flashcards

This version is the actively deployed and recommended implementation.

Refer to groq_version/README.md for full details.

---

🧠 Why Two Versions?

This repository is structured to highlight engineering decision-making, not just functionality.

Aspect	Gemini Version	Groq Version
Architecture	Cloud-heavy	Lean & portable
Cost	Paid cloud services	Free / minimal
Embeddings	Managed APIs	Local CPU-based
Goal	Feature completeness	Cost + deployability

This evolution mirrors real-world software engineering workflows.

---

🚀 Features (Common Across Versions)

• PDF upload and processing
• Text chunking
• Semantic search using vector similarity
• AI-powered summaries
• Flashcard generation
• Context-aware chat (RAG)

---

🏗️ Deployment

• Version 2 (Groq) is designed for deployment on Render
• Uses ephemeral storage safely
• Stateless architecture after embeddings are created

---

📌 Notes

• Secrets (.env) are intentionally excluded from version control
• IDE-specific files are ignored via .gitignore
• Both versions are preserved for educational and evaluative purposes

---

📖 Further Documentation

• See gemini_version/README.md for the cloud-based implementation
• See groq_version/README.md for the optimized, deployable implementation

---

✨ Final Thought

This project is not just about building an AI application —
it is about understanding trade-offs, scalability, cost, and deployment realities.

Both versions together tell the full engineering story.