The process behind AutoSAM: How I build with AI agents.
The process behind AutoSAM: How I build with AI agents.
AutoSAM Case Study
First, the honest part: AutoSAM is pre-launch, and there are no customers or revenue yet. It exists as a working, actively tested platform, and this page treats it as what it is for now: an engineering case study, with every number pulled directly from the codebase.
In December 2025 I founded GovCertix LLC and started building AutoSAM: a production-grade B2B compliance platform for defense supply chains.
Defense subcontractors inherit obligations that flow down from prime contracts through FAR, DFARS, and CFR clauses, and tracking what actually applies to whom is brutal, manual work. AutoSAM extracts those obligations from the federal regulatory corpus and turns them into something a supplier network can act on.
Built alone, end to end:
Product and architecture
Code and tests
The marketing surface
The first real decision was scope discipline. A solo founder's default failure mode is a thin demo of everything; I chose a production-grade build of a bounded core instead, which is why the test suite exists at all.
The second decision was the delivery system. Writing a platform this size by hand, alone, in under a year was not realistic, so I built a multi-agent orchestration setup: AI agents doing bounded implementation work inside a quality-gate system I designed. I rejected the alternative of unstructured AI-assisted coding early; without enforced gates, generated code rots faster than one person can review it.
Every change passes typed compilation, lint, architecture invariant tests, unit tests, and end-to-end tests before it lands
Stack: Next.js, React, TypeScript, Node, Postgres, Stripe
Extraction engine: a retrieval-augmented generation pipeline over the regulatory corpus
90 feature modules
2,000+ automated tests
As of August 2026, built solo
The obligation-extraction engine, the RAG pipeline, and the compliance workflows are real, working software, validated by the test suite rather than by a demo video.
The honest caveat repeats on purpose: it is pre-launch, and shipping to production is a different milestone from building.
90
Feature modules
2,000+
Automated tests
Solo
Built end to end, in under a year
The transferable asset is not the product, it is the delivery method. The orchestration and quality-gate system that made this build possible is the same system I bring to client work.
A second dividend: domain knowledge. Building an obligation engine is how I learned how flow-down requirements and NIST 800-171 evidence actually work in practice.
A discipline lesson: on a solo build, the test suite is the only teammate who never gets tired, so I invested in it like one.
