OCR and Embeddings Learning App

This repository is a learning project for Devscale AI-Enabled Python Batch IV. It demonstrates a simple OCR and embeddings workflow using Python, Mistral OCR, and ChromaDB.

What This Project Does

• Reads employee-handbook.pdf
• Sends the PDF to Mistral OCR
• Saves each OCR page as Markdown in ocr-output/
• Stores OCR text in a local ChromaDB collection
• Runs a semantic query against the stored document text

Tech Stack

• Python 3.13+
• Mistral AI OCR
• ChromaDB
• python-dotenv
• uv

Project Structure

.
├── app/
│   ├── main.py      # Query OCR text from ChromaDB
│   ├── ocr.py       # Run OCR and store page text embeddings
│   └── utils.py     # Shared Mistral and Chroma clients
├── chroma_db/       # Local ChromaDB persistence
├── ocr-output/      # OCR Markdown output per page
├── employee-handbook.pdf
├── pyproject.toml
└── uv.lock

Setup

Install dependencies:

uv sync

Create a .env file:

MISTRALAI_API_KEY=your_mistral_api_key

Usage

Run OCR and save the output into ChromaDB:

uv run python -m app.ocr

Query the embedded OCR content:

uv run python -m app.main

The current sample query asks:

What is the paternity leave policy

You can edit app/main.py to try different questions.

Learning Goals

This project is intended to help learners understand:

• How OCR turns PDF documents into machine-readable text
• How document text can be split and stored for retrieval
• How embeddings make semantic search possible
• How a local vector database can support document question answering