How We Built a Federal Contract Intelligence Platform in One Day
A real case study of building GovOpps AI Recon — an AI-powered SAM.gov opportunity discovery platform — from zero to live beta in a single session.
2026-03-28
Federal contracting is a $700 billion market. Most small businesses never touch it — not because they're unqualified, but because finding and evaluating the right opportunities is a full-time job.
We built a platform to change that. Here's exactly how we did it, what broke, and how we fixed it.
The Problem
SAM.gov lists every federal contracting opportunity. It's publicly accessible. But it's also overwhelming: thousands of contracts updated daily, each requiring hours of analysis to determine fit.
Small contractors — the ones who could actually win these contracts — don't have hours. They have one person doing three jobs.
The question: could we use AI to do the analysis automatically and surface only the opportunities worth pursuing?
What We Built
GovOpps AI Recon — a multi-tenant SaaS platform that:
- Connects to SAM.gov's public API and pulls opportunities matching your NAICS codes
- Scores each opportunity using Claude AI against your company profile (certifications, past performance, value range)
- Tags opportunities automatically (geography, competition level, contract type)
- Tracks your pipeline from discovery through proposal
The Tech Stack
- Next.js 15 with App Router and Edge Runtime (Cloudflare Pages compatibility)
- Supabase for multi-tenant auth, database, and row-level security
- Claude AI (Anthropic) for opportunity scoring — both Sonnet for precision and Haiku for batch operations
- SAM.gov API for live opportunity data
- Cloudflare Pages for deployment
What Broke (And What We Fixed)
Problem 1: Edge Runtime vs. Node.js APIs
Cloudflare Pages runs on Edge Runtime — a stripped-down JavaScript environment that doesn't support Node.js APIs like fs, path, or certain crypto methods.
Our first deploy crashed with three separate Edge Runtime errors in server action files.
Fix: We audited every server action for Node.js-specific patterns and replaced them:
- Stripped
revalidatePath(not supported in Edge) - Replaced inline server action closures with
.bind()pattern - Changed dynamic
import()inside closures to static imports at file top
Problem 2: Multi-Tenant Security
With multiple companies using the same platform, the risk of data leakage between tenants is real. We used Supabase Row-Level Security (RLS) policies to enforce tenant isolation at the database layer — every query is automatically scoped to the requesting organization.
Problem 3: AI Scoring Costs
Claude Sonnet is powerful but expensive at scale. For initial scoring of large batches, we built a two-tier system: Haiku for automatic background scoring (fast, cheap, good enough for initial ranking) and Sonnet for deep analysis when a user explicitly requests it.
The Result
Live beta with real users on day one. Zach Hodge, CEO of Hodge Group Investments & Contracting LLC, signed up during the session and immediately added his NAICS codes and ran a SAM.gov refresh.
It worked.
What This Demonstrates
You don't need a 6-month runway to ship production software in 2026. You need a clear problem, the right stack, and a disciplined build process.
AI handles the complexity that used to require a team. One developer with the right tools can ship what used to take three.
Gray Hodge is a Fractional Chief AI Officer and full-stack engineer. He builds AI-powered platforms for small businesses and government contractors. Work with Gray →
Gray Hodge is a Fractional Chief AI Officer and full-stack engineer. He builds AI-powered platforms for small businesses and government contractors. Work with Gray →