Healthcare AI·Hospital System (Panama)·4 months
MILA — Neonatal LLM Assistant
The Problem
Hospital staff couldn't consistently write clear, compassionate updates for NICU parents. Messages were either too technical, too brief, or missing critical policy references. Parents were left confused during the most stressful time of their lives.
The Approach
- 1Built a RAG pipeline over hospital policies and protocols using LangChain + Pinecone
- 2Created a simple clinician interface for drafting messages with automatic clarity checks
- 3Implemented role-based access with full audit trails for healthcare compliance
- 4Added a mandatory human approval step — no fully automated patient communication
The Results
42%
Faster update writing
97%
Policy reference accuracy
0
Factual errors in 300+ messages
82%
Staff adoption in month one
Tech Stack
PythonLangChainPineconeFastAPIOpenAIPostgreSQLNext.js
Key Lesson
“The biggest challenge wasn't the AI — it was earning trust from clinicians who were skeptical about AI touching patient communication. The mandatory human approval step was the key to adoption.”