Senior ML Engineer

IntelePeer is a healthcare-focused AI communications platform that powers AI voice agents and intelligent workflow automation. They are seeking a Senior ML Engineer to own the full lifecycle of machine learning systems, including designing training pipelines, fine-tuning models, and optimizing inference for production.

Responsibilities:

• Design, implement, and maintain end-to-end ML training pipelines — from raw data ingestion and preprocessing through model training, evaluation, and deployment
• Fine-tune large language models using techniques such as LoRA, QLoRA, and full fine-tuning; apply PEFT strategies to balance performance and compute cost
• Implement and experiment with reinforcement learning from human feedback (RLHF) workflows, including PPO (Proximal Policy Optimization) and GRPO (Group Relative Policy Optimization) for model alignment and preference optimization
• Host, serve, and optimize LLMs in production using inference frameworks such as vLLM, Text Generation Inference (TGI), Triton Inference Server, or ONNX Runtime
• Evaluate, benchmark, and select inference providers (e.g., Together AI, Fireworks, Groq, Replicate, AWS Bedrock, Azure OpenAI) based on latency, cost, throughput, and model capability trade-offs
• Build and maintain embedding pipelines — generate, index, and retrieve dense embeddings using vector databases (Pinecone, pgvector, Weaviate, or similar) for RAG and semantic search applications
• Implement and expose ML capabilities via Model Context Protocol (MCP) — enabling AI agents to call model-backed tools in a structured, context-aware manner
• Perform rigorous data analysis and processing: clean, transform, and curate datasets for training, fine-tuning, and evaluation; build data quality and validation pipelines
• Develop robust model evaluation frameworks — define metrics, build eval harnesses, run A/B experiments, and track regressions across model versions
• Collaborate with software engineers to integrate ML systems into product features via FastAPI services; ensure models are observable, versioned, and maintainable in production