Staff AI Engineer

Machinify is a leading healthcare intelligence company with expertise across the payment continuum, delivering unmatched value, transparency, and efficiency to health plan clients across the country. Deployed by over 85 health plans, including many of the top 20, and representing more than 270 million lives, Machinify brings together a fully configurable and content-rich, AI-powered platform along with best-in-class expertise. We’re constantly reimagining what’s possible in our industry, creating disruptively simple, powerfully clear ways to maximize financial outcomes and drive down healthcare costs.

We're hiring an AI Engineer to build LLM-powered products that actually ship. You'll live at the seam between research and production — turning models, prompts, and retrieval into reliable systems users depend on. Less "train a model from scratch," more "make foundation models do useful work at scale, cheaply, and without hallucinating."

What you'll do

• Design and build LLM applications: RAG pipelines, agents, tool-use systems, structured generation, multi-step workflows

• Own prompt engineering and evaluation — write evals before you write prompts, and treat both as code

• Integrate foundation models (Claude, GPT, Gemini, open-weight Llama/Qwen/Mistral) and know when to reach for which

• Build the unglamorous infrastructure that makes LLMs production-grade: caching, streaming, retries, fallbacks, cost controls, observability, guardrails

• Fine-tune or distill models when prompting hits a wall (LoRA/QLoRA, SFT, DPO)

• Partner with product and design on UX patterns specific to AI (latency masking, streaming UIs, confidence display, human-in-the-loop)

What we're looking for

• 3+ years of software engineering, with at least 1 year shipping LLM-based features in production

• Strong Python and TypeScript; comfortable in a real codebase, not just notebooks

• Deep practical knowledge of the LLM stack: prompting techniques, function calling, structured outputs, context management, embedding models, vector search

• Have built and maintained an eval suite — and have opinions on why most public benchmarks are misleading

• Comfortable with at least one orchestration approach (custom, LangGraph, Inngest, Temporal) and one vector store (pgvector, Pinecone, Turbopuffer, Weaviate)

• Cost- and latency-aware: can read a token bill and a trace and know where to cut

• Clear communicator who can push back when "just add AI" isn't the right answer

Nice to have

• Experience with agent frameworks and MCP (Model Context Protocol)