captureapplication.fairspec.org
Fairspec — data framework in TypeScript and Python
A specification for describing tabular datasets, with reference implementations in TypeScript and Python.
we author
Services & integration
AI engineering
Models behind an interface we control, with retrieval and evaluation around them.
- OpenRouter
- AI SDK
- Mastra
- MCP
- pgvector
What it is
Putting language models to work on a defined task — extraction, classification, matching, assistance — behind an interface with types, limits and a fallback.
Why we chose it
We route through OpenRouter so a model is a configuration value rather than a rewrite, and build on the AI SDK and Mastra for structured output, tool calling and agent state. Retrieval runs on pgvector in the database that already holds the data, which avoids operating a second store and keeps permissions in one place. MCP is how we expose our own systems to models, and the same discipline applies as anywhere else: a schema on the output, an evaluation on the result, and a deterministic path when the model is wrong.
Where we use it
In data normalisation and matching, where the alternative is an unmaintainable pile of rules, and in the tooling we build for our own engineering.
capturefireargs.dev
Fireargs — schema-first CLI builder
Schema-first CLI builder generating composable commands with built-in JSON I/O from Zod schemas.
we maintain
Get in touch
Tell us what you need built.
We are a small team in Portugal, working London hours. A site, an app, an internal tool, or the platform behind them — start with a sentence about the problem and we will take it from there.