A Tech Sight platform

Fynbos

The AI scaffold platform we build and operate ourselves. It is where the controls we recommend were written, tested, and in several cases learned the hard way.

What it does

Fynbos runs continuously in production. It is not a demonstration environment, and that distinction is the entire reason it is on this site.

Routes across providers

Work is dispatched to the model that suits it, with automatic fallback when a provider is unavailable. No single vendor can take the platform down.

Meters and caps its own spend

Per-task budgets, enforced in code rather than policy. An agent that would exceed its ceiling stops instead of asking forgiveness.

Gates its own changes

Nothing reaches production without a written specification, a review, and a recorded decision. The audit trail is a by-product of the process, not an export from it.

The platform runs the same discipline we sell. Every control in the assessment catalogue exists because it was needed here first.

Why a consultancy operates a platform

Because governance advice written by someone who has never carried the pager for an AI system tends to be correct and useless at the same time.

Running one means the recommendations have already survived contact with a real estate under load, and the failures that shaped them are ours rather than a client's.

Products built on it

  • LinguistPro, AI transcription for South Africa across thirteen languages, with multi-engine routing and speaker diarisation. Available now.
  • illusion-of-correctness, an open-source test-suite auditor. MIT licensed, on npm, and every check traced to a real production post-mortem.
  • TechSight Data Feeds, structured change data over a metered API. Built on the same platform and currently not offered for sale.

See what it would find in yours

The assessment applies the same controls to your estate that this platform runs on itself.

Book the two-hour diagnostic