Career
Forward Deployed Engineer at C3 AI
I lead delivery of production AI for federal and enterprise customers: technical lead and primary developer across a multi-program portfolio of U.S. Marine Corps work, three concurrent teams, and a run of internal systems that started as one engagement's problem and became reusable across other engagements.
Technical lead across Manpower & Reserve Affairs and the Marine Expeditionary Force personnel (G-1) and logistics (G-4) programs — workforce-planning and personnel-readiness systems spanning enlisted distribution, promotion and assignment planning, and force logistics. Led three concurrent teams of roughly nine engineers and held customer relationships end to end, from field users up to command-level leadership. Took MEF G-1 from a 108-page specification to production on the C3 AI Platform, deployed on Advana, the Department of War data platform operated by CDAO. Ran on-site workshops with stakeholders across the chain.
Delivered the blocking tail that took USDA's AI correspondence-management system to go-live. It runs an automated pipeline: ingest constituent mail, extract and flag PII, classify each item by topic, draft a response with an LLM, and route it through a multi-step approval workflow before sending. Owned server-side access scoping across agencies, a grid-fetch refactor, and a redaction pipeline that preserves the original document format, across roughly forty tickets and dozens of reviewed pull requests.
Shipped a translation system on a hard court deadline for an intellectual-property case: an OCR pipeline, a translator–critic–arbitrator loop that scores and reconciles output, and a 21-agent verification swarm running autonomously on a file-system inbox structure. It handled 1,081 documents and 2.2 million words in place of an outside vendor. Full write-up.
Joined a NATO contested-logistics program mid-stream to accelerate delivery: supply-network-risk and deployment planning with a rerouting optimizer, integrated with the alliance's logistics command-and-control system. Included on-site support in London.
Built a production harness that drives development asynchronously from a ticket to a reviewed pull request: it generates code, runs build, unit and end-to-end tests, and opens a PR per work unit behind an automated senior-tier review gate. Models are tiered — cheap for orchestration, frontier for the gates — holding the cost under $10 in model spend per work unit. Used it to auto-complete tickets and to power a multi-agent UI migration adopted by several teams.
Engineered a reusable method for verifying role-based access control, exercising both backend and interface as each role against the written permission matrix. It verified 176 role-by-capability combinations on a federal case-management system before go-live, and was packaged as a portable security-testing tool for other engagements.
Designed a data-ingestion framework that cuts onboarding a new external data source into a C3 application to roughly half a day of developer time, driven by a natural-language prompt and a config-driven connector setup. The interface is fully dynamic, so a new source needs little backend work, and the whole thing ships as a package any project can adopt.
Built a working prototype for a county law-enforcement agency in 48 hours on C3 Code: search-warrant generation matching the department's exact format, dynamic dashboards, geospatial mapping, PII redaction, and entity link analysis. Demonstrated live to county police chiefs.
Led delivery of C3's property-appraisal application across a run of Connecticut town assessors, the system appraisers use to produce accurate, defensible residential valuations at scale. Built the data-ingestion pipelines that onboarded each town, modeled the parcel and assessable-event data down to MLS field mapping and timed valuation history, and shipped the overview dashboards, IAAO metrics, mass-appraisal workflows, and automated-valuation-model output that appraisers work in daily.
Designed and deployed five production LLM agents spanning Sales, Finance, and Legal: two retrieval-augmented proposal-generation agents grounded in internal documents, a hypothesis-qualification agent, a sales-coaching agent, and an expense-analysis agent.
Led the AI SWAT team, C3's internal group building AI tooling across the company, and advised vice presidents and department heads on internal AI adoption. Also built a spec-to-design pipeline: a structured prompt distilled from subject-matter-expert interviews turns a program brief into a product spec, then an autonomous four-hour design run produces a full clicked-through application front end.
Sole customer-facing engineer on the company-wide platform-stability tiger team. Built crash-triage tooling that traces the system actions preceding a crash, led the field team's triage, and published internal guides on the methodology, as developer-impacting crashes fell by more than 90%. Also stood up the company's first pre-release platform testing program, building the framework and defining the quality standards with the CEO.