<aside> 💡

Important: Upload your work as a zip on this form at least 24 hours before the restitution.

</aside>

Instructions

🎯 Context & Objective

At Parallel, our agent analyzes the medical documents produced during a hospital stay to suggest the corresponding official diagnostic codes (CIM-10).

We provide you with a working V0 prototype in TypeScript. Initial feedback from partner hospitals has highlighted several inconsistencies and differences from medical coders' practices.

Your objective (~3–4 hours): Analyze the behavior of this prototype and implement at least one improvement to the agent.

⚙️ Environment & Model

• Starting point: GitHub repository
• Stack: TypeScript / Node.js.
• LLM of your choice: you may use a local model (e.g. Ollama or LM Studio with qwen3.5) or a personal API key for any provider. Your choice of model and its absolute performance are not part of the assessment.
• Pragmatic approach: we assess your analytical ability, critical thinking, and the quality of your engineering decisions—not the number of complex features you add.

📚 Materials Provided for the Exercise

In real-world settings, medical coders rely on CoCoA 2026. For this exercise, our V0 prototype uses a simplified version (coding-rules.md) that brings together PMSI principles, selected clinical criteria, and a restricted code set.

• 0. Primer : Le codage médical en France (PMSI & CIM-10)
• 1. Guide des règles métier — coding-rules.md
• 2. Dossiers médicaux synthétiques (4 séjours)
• 3. Historical reference labels (ground-truth.json)