Principal AI Engineer

Hearst•Dallas, TX

About The Position

Homecare Homebase is seeking a hands-on Principal AI Platform Engineer to build a greenfield AI software factory and serve as the senior technical authority for AI embedded in its products. The role is an individual-contributor position with no direct reports initially. The engineer will code daily, ship the first working agent platform, and enable engineering teams to build on it. The platform will run agents against the production codebase, with people writing specifications and validating results while agents handle much of the implementation. The environment primarily uses .NET, Azure, Azure DevOps, Kubernetes, Azure AI Foundry and MCP-based tooling, while the platform must support multiple languages and frameworks. The role is expected to deliver an adoptable agent template and shared framework, with at least one team using it on real tickets within the first 90 days. Success also includes telemetry for token spend, agent launches and tool-call traces; a baseline for measuring development-process impact; and a common standard across teams that lets each team maintain its own knowledge without building separate agents, models or tiers. The Principal AI Platform Engineer will design, develop and operate the agent platform; create developer-facing APIs, SDKs, MCP servers and extension points; engineer context and retrieval approaches for large codebases; and establish evaluation, safety, governance and observability practices appropriate to regulated environments. The role also partners with application teams, mentors engineers through working examples, shapes engineering standards and the AI roadmap with the CTO, and evaluates emerging AI capabilities, models and tooling. Applicants must be authorized to work in the U.S. without immigration sponsorship.

Requirements

Bachelor’s or advanced degree in Computer Science, Engineering or a related field, or equivalent practical experience.
At least 10 years of software engineering experience, with recent hands-on coding.
At least 3 years building and operating production AI systems in which agents call tools or change code; RAG or machine-learning products alone do not satisfy this requirement.
Proficiency in Python and modern AI tooling, such as LangChain, LangGraph, MCP, and OpenAI and Anthropic SDKs.
Hands-on experience building MCP servers or equivalent tool interfaces, including authentication and per-tenant access control.
Ability to make agents effective across multiple languages and technology stacks.
Hands-on experience deploying and operating AI workloads in cloud environments; Azure is preferred.
Solid understanding of LLMs, generative AI, agentic patterns and context engineering.
Proven RAG experience and judgment about when retrieval is preferable to curated context.
Experience with AI evaluation and observability, including quality measurement, cost and token management.
Working knowledge of handling regulated data such as PHI in AI systems.
Demonstrated experience mentoring engineers and raising technical capability across teams.
Provide evidence of personally built agent or MCP work, such as a repository, code sample or demo.
Must have the right to work in the U.S. without immigration sponsorship.

Nice To Haves

Experience integrating AI into C# and .NET applications.
Experience using AI coding tools such as Cursor or Claude Code on large codebases.
Familiarity with vector databases and retrieval tooling such as pgvector, Pinecone or Azure AI Search.
Exposure to LLMOps and evaluation tooling such as Azure AI Foundry, MLflow, LangSmith or CI/CD.
Familiarity with agent-to-agent communication standards such as A2A, and judgment about when a multi-agent design is warranted over a single agent with tools.
Experience with large legacy codebases and incremental modernization.
Interest in healthcare, EHR or revenue cycle applications.

Responsibilities

Design, write and operate an agent platform that can take work from specification to pull request against the production codebase, with human review.
Implement agentic patterns, orchestration and tool integration for engineering workflows.
Make the platform effective across the organization’s technology stack, including multiple languages and frameworks.
Define shared agent templates and a framework for engineering teams.
Own the AI lifecycle from concept through deployment, operation and iteration on Azure and Azure AI Foundry.
Shape how work is specified for agents and review outputs through the pull-request stage.
Partner with application teams to integrate AI into products and workflows.
Build APIs, SDKs, MCP servers and extension points for engineering teams.
Establish how MCP servers, tools and agent skills are packaged, versioned, discovered and distributed, including suitable access controls.
Design per-tenant data scoping so agents can access only the data of the agency they act for.
Design context, skills and retrieval patterns for agents working in large codebases, including an application-level knowledge map.
Determine when retrieval is appropriate and when curated context is preferable.
Establish versioning, review and governance for shared AI assets.
Build evaluation frameworks to measure output quality and detect regressions.
Define guardrails, failure handling and human-in-the-loop controls for regulated environments, including preventing PHI from entering agent context, prompts and logs.
Monitor and optimize AI systems for reliability, accuracy and cost efficiency, using measures such as token budgets, rate limits, retry caps and step limits.
Establish observability practices for supervised and autonomous AI usage.
Mentor engineers and increase AI engineering capability by building working examples in their codebases.
Define engineering standards and shape the AI technical roadmap with the CTO.
Evaluate emerging LLM, agentic AI, RAG and MCP capabilities and bring suitable approaches into production.
Advise on build, buy and partner decisions across the AI toolchain.
Maintain knowledge of commercial and open-weight models and match model tiers to tasks based on cost, latency, quality and data-handling terms.