AI/ML Engineering Intern — Healthcare LLM

About the role

AI is the moat that separates MedMinutes from legacy hospital information systems like Khanza or SIMGOS. Your work won't sit in a research notebook — every prompt, eval, and pipeline you build ships into production within weeks and gets used by real clinicians and hospital administrators.

You'll work across two flagship AI surfaces:

• Clinical Decision Support System (CDSS) — 4 production modules helping doctors with diagnosis verification, drug interactions, clinical guidelines, and BPJS claim verification. Each module has its own backend, prompt logic, and doctor-facing UI.
• BPJScan Smart QA — an LLM assistant that lets hospital admins ask free-form questions over millions of rows of BPJS insurance claim data ("which DRGs are we under-coding?", "show me readmissions in the last 30 days").

What you'll do

• Build, evaluate, and iterate on LLM prompts (Claude, Gemini, Groq, OpenAI) for clinical and claims use cases
• Design eval frameworks — measure accuracy, hallucination rate, latency, and cost
• Implement RAG pipelines: embeddings, vector search, reranking, grounding
• Ship features to production every 1–2 weeks and watch real doctors use them
• Optimize prompts for cost via prompt caching, batching, and model routing
• Collaborate with the product team to scope new AI modules from clinical pain points
• Write technical documentation and present results to engineering and clinical stakeholders

You're a fit if you

• Are studying or recently graduated in CS, Data Science, AI/ML, or a related field
• Are comfortable with Python, Git, and REST APIs
• Have shipped at least one project using LLM APIs (OpenAI, Claude, Gemini, etc.) — coursework, hackathon, or personal project all count
• Understand basic ML concepts: embeddings, fine-tuning vs prompting, eval metrics
• Are curious about healthcare and how AI can actually help — not just the hype
• Can read English technical docs and understand Bahasa Indonesia clinical context

Bonus points

• Experience with FastAPI, Next.js, or TypeScript
• Familiarity with vector databases (ChromaDB, Pinecone, Weaviate, Qdrant)
• Past projects in NLP, RAG, agent frameworks, or fine-tuning
• Experience with Anthropic SDK (prompt caching, tool use), Groq, or Vertex AI
• Comfort reading clinical or insurance documentation (BPJS, ICD-10, INA-CBG)

Sample projects you might tackle

• Build a regression eval set for the CDSS drug interaction module — track accuracy as we change prompts/models
• Design a RAG pipeline over 502K+ rows of BPJS claim data with sub-2-second latency
• Reduce monthly LLM cost by 40%+ via prompt caching, batch API, and smaller-model routing
• Add tool-use / agentic behavior to clinical assistants (lab lookups, drug DB queries)
• Improve a CDSS module's accuracy from baseline to a measurable target

What we offer

• Real production AI deployments — your prompts run in real hospitals every day
• Weekly 2-hour mentoring sessions with the CTO and senior engineers
• Direct access to existing AI infra: Anthropic, Gemini, Groq, OpenAI, ChromaDB, Vertex AI
• Hybrid work — Semarang office or fully remote, your choice
• Stipend (negotiable based on experience and scope)
• Letter of recommendation, project portfolio piece, and demo video for your CV

Project timeline

4–6 months. We work in 2-week iterations: pick a problem, design eval, implement, measure, ship. By the end you'll have shipped at least 8–12 production-grade AI improvements and a final case study suitable for portfolio or thesis.

Location & format

Hybrid. The team is based at our Semarang office (Jl. Gunung Sawo No. 17), but full remote is fine — we only ask for one weekly mentoring sync and presence at milestone demos. Start date is flexible.