OpenAI API

I use the OpenAI API in two ways: directly, for structured analysis, and as a compatible interface to other providers.

Directly:

• ScopeGuard AI calls the Responses API with strict JSON Schema output to compare dated client sources against an agreed baseline. Every finding cites source IDs, and a second Zod pass rejects any citation that doesn't exist. We built it for OpenAI Build Week, and Codex did most of the implementation.
• CivicAI Readiness Blueprint adds an optional, advisory policy insight in a strict schema. The scoring itself is local and deterministic, so the app still works when the API doesn't.

Through OpenAI-compatible endpoints:

• QuotePilot AI and KhidmatConnect AI call Qwen this way.
• CaptionForge AI calls Fireworks vision and text models.
• NeighborOps AI runs a Strands agent on free OpenRouter models.

The rule is the same everywhere: model output is a proposal. It's validated against a schema, labelled with its source, and kept away from anything irreversible until code or a person has checked it.

Evidence level: strong, from several shipped projects.