Building a Document Processing Workflow in 30 Minutes

Mistral's Workflows is an orchestration platform for building, running, and monitoring complex AI-powered workflows. Built on distributed systems infrastructure, it offers durability and fault-tolerant features, and supports a developer-friendly SDK.

This tutorial covers combining Mistral OCR, Agents, and Workflows to build a medical document processing pipeline. It takes PDF files such as prescriptions or hospital receipts as input and goes through the following 3 steps.

• Mistral OCR: Extract raw text from PDF
• AI Agent (Classification): Classify document type and produce a confidence score
• AI Agent (Extraction): Extract patient information and specific fields into structured JSON data

In particular, the Human-in-the-loop feature allows the process to pause and wait for user review before proceeding to the next step if the classification confidence is below a set threshold. This makes it possible to reliably manage complex, long-running processes that are difficult to implement with simple asynchronous queues or API chains alone.

The build process uses Python 3.12, uv, and Streamlit, and covers the entire process from project initialization to frontend implementation and load testing.