Technology Strategy Director

At Freddie Mac, our mission of Making Home Possible is what motivates us, and it’s at the core of everything we do. Since our charter in 1970, we have made home possible for more than 90 million families across the country. Join an organization where your work contributes to a greater purpose.
Position Overview:
Freddie Mac is seeking a hands-on AI engineering leader to serve as Director within the Single-Family Acquisitions (SFA) division, reporting to the Sr. Director of GenAI Engineering & Enterprise Rules Platform. This role owns the technical core of SFA’s AI Systems Solutioning, including Agent Engineering & Design. This includes (1) the overall design and execution of AI Systems development from Ideation & Business casing to Specification creation & execution through deployment, (2) the evaluation harnesses and graded assessment sets that determine whether an agent or AI System is ready for production, and (3) the operating layer that keeps agents accountable after go-live. The role is approximately 70% hands-on architecture and code and 30% leadership of a GenAI engineering efforts and platform and vendor delivery partners.
The Director is forward-deployed by design. The role partners with the divisional GenAI PMO, Enterprise Operations & Technology (EO&T), and Digital Experience, and works alongside the Enterprise Business Rules platform team so that agent-driven and rule-driven decisioning share a common domain vocabulary, evaluation approach, and traceability model.
Our Impact
Freddie Mac was established by Congress to provide liquidity, stability, and affordability to the U.S. housing market. Within Single-Family Acquisitions, our technology platforms and decisioning systems are foundational to ensuring loans sold to Freddie Mac meet eligibility, credit risk, and regulatory requirements at scale.
Your Impact
This role is a player / coach model. You will establish the engineering discipline that lets SFA move agentic workflows from pilot to production and keep them there: durable orchestration, accuracy that is measured rather than asserted, and operational transparency that satisfies both business owners and model risk oversight. Then use it. The capabilities built here—retrieval, orchestration, evaluation, and monitoring—become the shared substrate that every SFA use case inherits, so each delivery makes the next one faster. Building knowledge graphs that represent how the business functions and designing their use, matters.
Leadership
Lead GenAI engineering efforts, execution, and delivery partners under a single delivery model, while remaining the senior hands-on engineer
Stand up a forward-deployed engagement model (build, consult, advise, monitor) that embeds AI engineers alongside SFA business teams and is accountable both for what reaches production and for how it behaves afterward
Institutionalize spec-driven development as the team’s way of working: specifications, acceptance criteria, and evaluation targets are written before build loops begin, so every capability is auditable from intent through implementation
Establish and maintain the team’s pattern library—reusable kickstarters, skills, and reference implementations—built daily with AI coding assistants (GitHub Copilot, Amazon Kiro, OpenCode) under the spec-driven workflow
Partner with the divisional GenAI PMO on capability roadmaps, prioritization, and enablement; with EO&T on reference architectures and platform standards; and with Digital on product ownership and delivery practices
Mentor engineers on agent design, evaluation discipline, and production operations; serve as a credible technical voice with senior leadership on agentic readiness
Transformation & Delivery
Architect and deliver a durable agentic workflow platform: a workflow engine with persistent state, declarative agent configuration so new agents onboard without engine changes, tool and API integration, and agent interoperability (MCP, agent-to-agent handoffs)
Design a shared domain graph—a unified knowledge and service layer—as the common vocabulary across agents and rules, enabling validation across workstreams without coupling them to each other
Build the assessment harnesses and graded datasets that gate promotion: scored runs, full trace capture, and regression checks; use the same harnesses to drive prompt iteration and model selection across providers
Establish hybrid retrieval (vector plus graph) over a governed document substrate—Seller/Servicer Guide, policy, and procedural content—with docs-as-code ingestion and approvals, so every new use case inherits ingestion, guardrails, and observability
Engineer the post-launch operating layer: trace analysis linking agent decisions back to source documents and prompt versions, token and cost telemetry per agent run, SLO dashboards (latency, cost, throughput), drift detection, and handoff packages (runbooks, monitoring, documentation) so business owners can run what was built
Personally deliver the initial production use cases in credit risk and quality control end to end, from architecture through implementation, operations, and iteration
Technology & Engineering Practice
Agent systems: multi-agent orchestration on LangGraph and LangGraph Platform (AutoGen or equivalent), durable workflow state, declarative agent configuration, tool and API integration, agent interoperability (MCP, agent-to-agent handoffs), and vendor SDKs and agent development kits (AWS Bedrock, OpenAI, Anthropic)
Retrieval and knowledge: hybrid RAG and Graph RAG, Neo4j knowledge graphs, vector search (OpenSearch or equivalent), shared S3 document substrates, and docs-as-code corpora with governed approvals
Evaluation and operations: LangSmith for tracing, assessment datasets, and experiment tracking; graded assessment harnesses and hill-climbing improvement loops; release gates; trace lineage; token and cost telemetry; SLO dashboards; drift detection
Engineering practice: spec-driven development, production Python, TypeScript/React, AI coding assistants, AWS and Azure, security architecture, and governance artifacts for regulated release gates
Governance, Risk, & Compliance
Embed Responsible AI engineering across all solutions: content guardrails, adversarial red-teaming, bias and fairness reviews, model validation, and the governance artifacts that clear release gates
Ensure explainability and accountability: every agent decision traceable to its source documents, prompt version, and evaluation evidence
Ensure compliance with model risk management, data protection, and regulatory requirements applicable to GSE decisioning
Qualifications
12+ years building and operating production ML and AI systems in regulated environments, with 5+ years leading engineering teams
Demonstrated delivery of a multi-agent orchestration platform to production as a hands-on engineer on LangGraph (or equivalent), including durable workflow state and declarative agent configuration
Experience building evaluation harnesses, graded datasets, and release gates for LLM systems, and operating them in LangSmith (or equivalent tracing and observability tooling)
Knowledge graph and hybrid RAG experience at production scale with Neo4j, vector search, and governed document corpora
Track record of spec-driven development: writing the specification and acceptance criteria first, then building with AI coding assistants against them
Production Python; TypeScript/React a plus; AWS (Bedrock) or Azure
College degree or equivalent experience; advanced studies preferred
Financial services or mortgage domain experience strongly preferred
Keys to Success in This Role
Forward-deployed instinct: comfortable embedding with business teams and staying in the code
Insists on demonstrating value of AI Systems delivered
Plans before building: treats the spec and the evaluation set as first-class deliverables, not afterthoughts
Systems thinker who designs for the next use case, not just the current one
Clear communicator with engineers, business owners, and executives
First 12–18 months, success criteria includes:
Agentic workflow platform in production with at least two SFA use cases (credit risk, quality control) running on shared orchestration
Evaluation harness and release gates adopted as the standard for all SFA agent promotions
Observability layer live—trace lineage, cost telemetry, SLO dashboards—with runbooks handed to business owners
Shared domain graph in use by both agentic workflows and the Business Rules platform
Measurable reduction in time-to-production for new agentic capabilities
Current Freddie Mac employees please apply through the internal career site.
We consider all applicants for all positions without regard to gender, race, color, religion, national origin, age, marital status, veteran status, sexual orientation, gender identity/expression, physical and mental disability, pregnancy, ethnicity, genetic information or any other prote