One Afternoon, Seven APIs, One Answer: The Composable AI Tool Pattern

The most useful AI tools we build don't start from scratch. They start from public data that already exists, scattered across systems that don't talk to each other, and they use AI to pull it together into something actionable.

We call this the composable AI tool pattern. Here's a concrete example of it in action.

The problem

A utilities company needed a way for their customer service reps to quickly validate what type of business is at any given service address. The business type — classified by a NAICS code — determines power restoration priority after outages. The data in their system was stale. A tool to fix it didn't exist.

The pattern

Instead of building a new database or data collection process, we mapped the question — "what type of business is this?" — to every public source that might already have the answer:

• Healthcare providers → NPPES registry (CMS)
• Federal contractors → SAM.gov
• Public companies → SEC EDGAR
• Nursing homes and dialysis centers → CMS Care Compare
• Emergency services, airports, broadcast stations → OpenStreetMap
• Federal award recipients → USASpending.gov
• Everything else → web scrape of the company's own website

Seven sources. All public. All free. All queryable by API.

The AI layer — Claude Sonnet — receives all seven results simultaneously and synthesizes them into a single answer with a confidence rating and source citation. When multiple sources agree, confidence is High. The explanation tells the user exactly which sources confirmed the classification and whether they agreed.

Why this pattern works

The insight is that most "we need to know X about a business" problems are already partially answered in public registries — you just need to query the right ones and reconcile the results. The AI doesn't replace the data sources. It replaces the analyst who would otherwise spend 20 minutes manually checking each one.

The cost model is what makes it commercially interesting. Seven APIs, all free. Claude API at this volume costs pennies per query. Cloudflare Functions hosting is effectively free at this scale. A tool that would have required a data vendor contract and an analyst team a few years ago now runs for a few dollars a month.

What it means for operational AI

The composable pattern is replicable across dozens of operational problems:

• Contractor license verification (state licensing boards + SAM.gov + NPPES)
• Business credit risk (SEC + USASpending + state registrations)
• Healthcare provider credentialing (NPPES + CMS + state medical boards)
• Property classification (county assessor APIs + OpenStreetMap + business registries)

The architecture is the same in every case: identify the public registries that hold relevant data, build a parallel query layer, pass results to AI for synthesis with confidence scoring. The specific sources change. The pattern doesn't.

This is what we're building toward at Noevant: infrastructure that makes the composable AI pattern accessible without requiring you to build each piece from scratch. The tool we built for this client is a proof of concept. The operational AI stack is the product.

Noevant is the commercialization vehicle for JARVIS — the AI operational stack built and validated across six companies by 2057 Holdings.

Technical deep dive: Building an AI NAICS Code Lookup Tool by Jesse Myers.

Business case: We Built an AI-Powered NAICS Code Lookup Tool in an Afternoon — 2057 Holdings.

Enterprise data strategy: The NAICS Problem Is a Symptom — Safire Business Services.