Work

What was in the way, and what changed.

Every project begins with a problem the business couldn't solve with the tools it had. Here are two, and the systems built to solve them.

Projects

Multi-location healthcare groupOne view over three systemsAI in their own AWS account

One view of the numbers for a healthcare group and its locations

Situation

A healthcare group with several locations reported by hand from three separate systems. Leadership saw the numbers weeks late; location managers saw spreadsheets.

Stakes

Protected patient information. Each location may see only its own patients; the group must see all. Any tool that put patient data in a vendor's account, or in a chatbot, was out of the question.

What was built, and why in this order

The dashboards are designed first on summary reports with no patient information, so the layouts and definitions are agreed before anything sensitive is touched, with small counts hidden so nobody can be identified. The live systems connect inside the group's own AWS account, under the group's own AWS agreements, with one database in which every record is labelled by location and the database itself enforces who sees what. The AI that writes the narrative summaries runs through Amazon Bedrock inside AWS in the group's region, under terms that keep patient data from being trained on or passed anywhere.

What the system does

  • Gives leadership every location's numbers the same morning, instead of weeks later
  • Shows each location only its own patients, enforced in the database rather than trusted to the dashboard
  • Replaces manual report assembly; the hours it takes are measured before and after
  • Keeps patient data in the group's own AWS account; the AI that writes the summaries runs through Amazon Bedrock inside AWS, in the group's region
  • Leaves the group owning the code, the data, the AWS account and the documentation
Consumer forecast appA product, built and run

A snow forecast that shows its work

Situation

Skiers planning a trip get a single number and no sense of how sure it is, and the number can change from one day to the next.

Stakes

Trust. A forecast that shows five models, the spread between them and how it verified last week is one people come back to. It also had to run unattended.

What was built

A worldwide resort database, a service that blends five weather models hourly and converts precipitation to snow the way forecasters do, nightly verification against snow-station observations with per-resort correction, and a phone-first web app with a native app to follow.

What the system does

  • One forecast from five models, with the spread shown instead of hidden
  • Checks itself every night against what actually fell, and corrects per resort
  • Covers about 4,900 resorts worldwide from a single data pipeline
  • Refreshes itself and runs unattended on a small footprint
The same discipline applies to every build: documented, monitored, and designed to be run by whoever comes next.

Tell us the problem

What is in the way at your business? Describe it and which systems are involved; you'll hear whether Birdrock can help.