← All work

Data infrastructure

Autonomous acquisition

A self-healing adapter fleet reading 800+ independent sources nightly.

The system

The system ingests structured schedules from hundreds of sources that were never designed to be read by machines. Thirty-five adapter families cover the major platforms; a browser-driving fallback handles the long tail. The whole sweep runs unattended, every night.

When a source changes its markup, the adapter notices, repairs itself, and files the diff — maintenance is the product, so maintenance is automated.

The hard parts

No APIs exist

Of the long-tail sources sampled, zero exposed structured data. The fleet drives real browsers where it must and raw HTTP where it can.

Self-healing

Adapters detect schema drift and regenerate their own selectors; a model assist is reserved for the stubborn few percent.

Unattended by design

The full sweep runs nightly on a scheduler at 6× concurrency and completes in fourteen minutes.

In numbers

861
independent sources
35
adapter families
65,458
live records
14 min
full nightly sweep

TypeScript · headless browsers · Firebase · scheduled functions

Happy to walk through this one properly — what it does, how it's put together, and what it would take to build something like it for you.

Vague is fine.

What kind of help? optional — pick any