AI/ML lead (on-site, no remote, Astana)
Summary
Lead the design and deployment of an AI-powered healthcare platform that reduces patient readmissions using multi-step LLM agents, RAG systems, and fine-tuned open-source models.
Зарплата: 4000–5000 USD (до вычета налогов)
MDiet is an early-stage HealthTech startup developing an AI-powered platform that helps reduce patient readmissions.The product targets the US healthcare market.
Relocation to the US upon successful company growth (discussed individually)
Tech Stack:
- LangGraph + LangChain (Agentic AI)
- vLLM + Hugging Face
- PyTorch + PEFT (LoRA/QLoRA)
- Vector databases (pgvector, Qdrant, Weaviate, Chroma, Pinecone)
- Advanced RAG + Prompt Engineering + Embeddings
- FastAPI + PostgreSQL
- Docker + Kubernetes + async processing (Arq / Celery / Redis Streams)
- Apache Airflow (data pipelines)
- AWS Cloud (EC2, S3, SageMaker, Lambda, RDS, Bedrock)
- Structured output (Outlines / Guidance / Instructor)
- Evaluation & observability (Langfuse, RAGAS / DeepEval, Prometheus + Grafana)
- MLOps practices
II + 2- two=?
Responsibilities:
- Design and build multi-step LLM agents and pipelines for medical data curation
- Develop production RAG systems with high grounding and citation quality
- Deploy and optimize self-hosted LLM inference (vLLM) with cost and throughput optimization
- Fine-tune open-source models on domain-specific medical data with measurable improvement
- Build data pipelines and versioning for documents, embeddings and training data
- Create evaluation frameworks, guardrails and hallucination control systems
Implement reliable structured output and data extraction
- Work closely with physicians and Platform Architect
Requirements:
- Strong hands-on experience with production LLM systems (RAG + agents + inference)
- Experience with fine-tuning open-source LLMs (LoRA/QLoRA/PEFT) with measurable results
- Experience building data pipelines and versioning for LLM/RAG workloads
- Strong skills in evaluation, hallucination reduction and structured output
- Production Python + FastAPI + async processing
- Experience with self-hosted inference (vLLM or similar) and GPU optimization
- English proficiency at C1 level (strong written and spoken communication skills)
- Bachelor’s degree (or higher) in Computer Science, Artificial Intelligence, Machine Learning, Data Science, or a related technical field
- 5+ years in ML/AI, with 1.5–2+ years in production LLM projects
- Experience with MLOps
Nice to have:
- Experience in medical / clinical NLP or regulated domains
- Cost modeling and optimization of LLM workloads
- MLOps and production fine-tuning pipelines
- Apache Airflow, Prometheus/Grafana, or AI Solution Architecture
- Publications in relevant fields