AI Engineer
CodeRound is hiring for this role.
- Senior AI Engineer
- Senior ML Engineer
- AI/ML Engineer
- Applied AI
- Machine Learning
- Deep Learning
- Python
- PyTorch
- Hugging Face
- LLMs
- Generative AI
- Open Source LLMs
- Llama
- Qwen
- MedGemma
- Self-Hosted LLMs
- LLM Inference
- vLLM
- SGLang
- TensorRT-LLM
- GPU Optimization
- Model Serving
- Model Fine-Tuning
- SFT
- LoRA
- QLoRA
- DPO
- NLP
- RAG
- Retrieval Augmented Generation
- Embeddings
- Vector Databases
- Hybrid Search
- BM25
- Reranking
- Knowledge Graphs
- Neo4j
- ArangoDB
- Graph Retrieval
- LLM Evaluation
- LLM Observability
- Langfuse
- LangSmith
- Arize
- Braintrust
- LLM Monitoring
- MLOps
- LLMOps
- Prompt Engineering
- Structured Outputs
- Function Calling
- MCP
- AI Agents
- Agentic Workflows
- Multi-Agent Systems
- Human-in-the-Loop
- Healthcare AI
- Clinical NLP
- Revenue Cycle Management
- Medical Coding
- Claims Processing
- ICD-10
- CPT
- HCPCS
- HCC
- NCCI
This role is for one of our client companies — a well-funded AI-driven healthcare technology startup transforming revenue cycle management through advanced AI. The company is building a production-grade healthcare AI platform that automates complex revenue cycle workflows, including medical coding, claim edits, denial management, and appeals. As a Senior AI/ML Engineer, you will own the applied LLM and retrieval infrastructure, building scalable, auditable, and cost-efficient AI systems using self-hosted open-source models, knowledge graphs, and advanced evaluation frameworks.
What you'll do
- Self-Hosted LLM Infrastructure: Deploy, fine-tune, and operate open-source LLMs as the primary inference stack, optimizing GPU utilization, throughput, latency, batching, and inference costs.
- Model Fine-Tuning: Own end-to-end fine-tuning workflows using SFT, LoRA, QLoRA, and DPO to adapt models to clinical notes, claims, and payer-specific data.
- Knowledge Graph Engineering: Design and maintain knowledge graphs representing medical coding standards, healthcare policies, payer rules, and relationships across coding systems.
- Retrieval Systems: Build and optimize embedding-based retrieval pipelines using chunking, embedding models, hybrid search, vector databases, and reranking.
- Grounded AI Systems: Combine graph traversal and dense retrieval to ensure coding recommendations, claim edits, and appeal responses are backed by reliable and auditable evidence.
- Evaluation Infrastructure: Build continuous evaluation pipelines and offline benchmarks using human-validated datasets to assess model accuracy, retrieval quality, and system reliability.
- Production Monitoring: Monitor hallucinations, model drift, regressions, and output quality through shadow deployments, cohort-level accuracy tracking, and production observability.
- LLM Application Development: Design prompts, context pipelines, structured outputs, and function-calling workflows for medical coding, denial classification, claim edits, and appeal generation.
- Agentic Workflows: Build MCP servers and multi-step agent workflows that integrate internal tools, coding lookups, payer-rule checks, and denial classification systems.
- Reliability & Human Oversight: Establish deterministic versus LLM-driven execution boundaries and implement human-in-the-loop checkpoints for critical healthcare workflows.
- Performance Optimization: Track and improve coding accuracy, clean-claim rates, denial impact, cost per chart, and end-to-end system latency.
- Cross-Functional Collaboration: Work with product, engineering, and healthcare domain teams to translate complex business requirements into reliable production AI solutions.
Must have
4+ years of experience in AI/ML Engineering, Machine Learning, or Applied AI, including at least 6 months of hands-on production LLM experience. Strong proficiency in Python, PyTorch, and Hugging Face Transformers. Hands-on experience deploying, serving, and operating self-hosted open-source LLMs using vLLM, SGLang, TensorRT-LLM, or equivalent inference frameworks. Strong understanding of LLM inference optimization, including GPU utilization, batching, KV-cache management, throughput, latency, and serving costs. Experience designing embedding-based retrieval systems and/or knowledge graphs for grounded LLM applications. Strong understanding of RAG architectures, including document chunking, embedding generation, hybrid search, retrieval optimization, and reranking. Demonstrated experience building LLM evaluation infrastructure, including offline benchmarks, model evaluation, regression testing, and production monitoring. Experience developing production-grade AI systems with monitoring, incident management, performance optimization, and end-to-end ownership. Strong understanding of prompt engineering, structured LLM outputs, function calling, and integrating LLMs into complex application workflows. Ability to design reliable, scalable, and auditable AI systems with a focus on production performance and cost efficiency.
Good to have
Hands-on experience with fine-tuning techniques such as Supervised Fine-Tuning (SFT), LoRA, QLoRA, and DPO. Experience working with open-source models such as Llama, Qwen, MedGemma, or similar architectures. Familiarity with graph databases such as Neo4j, ArangoDB, or equivalent, including graph-based retrieval. Experience with vector databases, BM25, dense retrieval, hybrid search, and reranking models. Exposure to LLM observability and evaluation tools such as Langfuse, LangSmith, Arize, or Braintrust. Experience building MCP servers, tool integrations, or agent orchestration frameworks. Familiarity with multi-agent workflows, deterministic tool execution, and human-in-the-loop systems. Experience working with healthcare AI, Revenue Cycle Management (RCM), medical coding, claims processing, or other regulated domains. Understanding of healthcare coding standards and policies such as ICD-10-CM, CPT, HCPCS, HCC, and NCCI. Experience building AI systems with strict requirements around traceability, explainability, and auditability