felixnatanaelbutarbutar/dokter-penjaga — reverse-engineered prompt

Build me a safety first medical assistant called Dokter Penjaga. I want a simple web chat and a REST API where people can ask health questions in Indonesian or English, but it should act more like a cautious medical helper than a normal chatbot.

The most important part is safety. If the user mentions emergency symptoms like chest pain or trouble breathing, skip the usual AI answer and immediately show urgent guidance to seek emergency help. Before anything goes to the model, hide personal details like names, ID numbers, and phone numbers. Also block obvious prompt attacks and any requests for harmful or dangerous dosing advice. For normal questions, answer using a small medical knowledge base with citations or source info, and prefer newer documents when possible.

Keep the frontend minimal but usable, and include a health check plus API docs. Since this is an MVP, it is fine to use structured sample medical data instead of full PDF ingestion, but make the pipeline ready for that later. Please also wire up the evaluation scripts so I can test triage, retrieval, factuality, and guardrails. Look up current docs online if you need to.