We rebuilt the U.S. military's enlistment system. Now we're bringing that experience to the rest of government.

Half the states still process unemployment insurance claims on COBOL mainframes. The IRS Individual Master File is sixty years old and now scheduled for modernization completion in 2028. Treasury runs assembler. State Medicaid eligibility, child welfare case management, DMV title and registration, and corrections platforms are routinely twenty to forty years old. These systems carry the actual mission, every single day. They also carry the workforce risk: the engineers who maintained COBOL and Assembly are retiring or already gone.

We've done this work at federal scale. Over three years, we replaced the U.S. military's enlistment processing system, MIRS, for the U.S. Department of Defense. Twenty-five years old, eleven external integrations across all six branches, deployed at sixty-five Military Entrance Processing Stations nationwide, processing every applicant who joins the U.S. armed forces. Modern React front-end, Ruby on Rails back-end, GraphQL API layer replacing text-file integrations, AWS infrastructure. Onboarding time for new field advisors went from months to days. Millions of applicants processed and 90,000 enlistment contracts signed. Gold Stevie Award (2024).

The pattern transfers. State unemployment insurance, Medicaid eligibility, public assistance, child welfare, licensing and permitting platforms, and small-to-mid federal agency systems all share the same shape: a working but outdated core, dozens of integrations, multiple stakeholder constituencies, and zero tolerance for downtime during the cutover. We do this without the $50M prime contract.

Caseworkers at state agencies routinely sit at the intersection of three or four disconnected systems: an eligibility engine, a document repository, a notes system, and a separate platform for mailing notices to claimants. Application data gets re-keyed. Case status is tracked in spreadsheets. Citizens wait weeks because nothing is integrated. The pandemic made this visible at every state UI office; it's still the everyday reality at Medicaid, SNAP, TANF, child welfare, and most state-licensed-occupation boards.

The technology pattern is well-understood: a unified case record across systems, document ingestion with structured extraction, status workflows that reflect the real adjudication process, and AI-assisted document review for caseworkers handling thousands of pages of medical records, tax filings, or licensing documentation. The VA's Automated Decision Support tool dropped average disability claim processing time from 141 days to 81 days, a 42% reduction in less than a year, while clearing the backlog below 100,000 for the first time since 2020.

For state and county agencies, this is where the engagement starts. Not "let's replace the entire eligibility system in three years." Instead: identify the highest-volume case type, build the unified caseworker view, layer in document automation, and let the platform earn its expansion.

Pennsylvania ran a one-year ChatGPT Enterprise pilot with 175 state employees across 14 agencies. Result: an average of 95 minutes saved per user per day, with 85% reporting a positive experience. Governor Shapiro expanded the program to 3,000 employees in October 2025. The VA onboarded 95,000 employees to its internal generative AI platform; users report 2-3 hours saved per week and 70%+ higher job satisfaction. Federal AI use cases jumped 105% year-over-year, with Microsoft Copilot deployed in 102 reported use cases and Anthropic's Claude in 25.

The bottleneck isn't the AI tools anymore. GSA's OneGov agreements (August 2025) make Claude and ChatGPT essentially free at $1 per agency per year. The bottleneck is integrating those tools into the workflows that caseworkers, claims examiners, environmental inspectors, attorneys, and program staff actually use every day. That requires understanding the work, the regulatory constraints, and the data the AI needs access to. None of which the foundation model vendor will do for you.

We build that integration layer. Document review tools for benefit examiners with redaction-aware retrieval. Drafting assistants for caseworkers writing notices and determinations. Search and summarization across decades of agency policy and prior decisions. The tools that turn 95 minutes saved per day into agency-wide reality, deployed inside your security boundary, with audit trails your IG can defend.

Every public sector technology leader has the same uncomfortable conversation with their leadership: "How many people did we serve this quarter?" And the answer requires pulling data from six different systems, three of which can't talk to each other directly, two of which only export PDFs, and one of which the vendor went out of business in 2017. Federal agencies face this across program offices. States face it across counties. Counties face it across special districts. Multi-state interstate compacts face it across jurisdictions.

For the MEPCOM rebuild, we consolidated requirements across more than a dozen stakeholders and replaced eleven legacy text-file integrations with a single GraphQL API layer that serves all six military branches plus external HR, medical, and verification systems. That same pattern, an integration layer that doesn't require ripping out the underlying systems, works for state child welfare data sharing across counties, federal grant reporting across program offices, public health data exchange across jurisdictions, and the cross-agency dashboards your leadership keeps asking for.

The work isn't glamorous and it doesn't make conference keynotes. It's the data foundation that everything else depends on, including the AI initiatives leadership keeps asking about.

CMMC 2.0 enforcement began November 10, 2025 for DoD contractors and their subs. As of late 2025, only about 0.5% of the roughly 80,000 organizations DoD estimates need Level 2 certification have actually achieved it. CISA's Zero Trust Maturity Model is now the federal expectation, with FY 2025 IG FISMA Reporting Metrics adding ZTA-specific maturity measures. GovRAMP (formerly StateRAMP) has expanded to 33+ states and 60+ cities. The DOJ ADA Title II rule requires accessibility compliance for state and local government websites for entities serving over 50,000 by April 2026.

These aren't checkbox exercises. CMMC requires architectural changes. Zero trust requires identity and segmentation work that often touches the application layer. ADA accessibility requires real engineering at the design system, component, and content level. We've worked at IL5 with controlled unclassified information for the Department of Defense, built compliance-aware architectures for utilities under NERC CIP, and shipped accessible, WCAG-compliant interfaces for systems that serve the entire U.S. military enlistment population.

The work fits naturally inside modernization, rather than as a separate compliance project. Architectural decisions, data access patterns, identity integration, and accessibility get baked in from the first sprint, not bolted on after launch.