The problem
I worked on a UK government programme with over 200 people in more than 15 teams, called pods. Information about who was in each team was spread across spreadsheets and people’s memories.
No one had a clear, accurate view of:
- who was in which team
- which teams owned which cloud accounts
- how a person’s team decided what they could access
When there was an incident, it took time to work out who to call. It was hard to work out which team should pay for what. New starters took days to get set up.
What I built
I built a tool that draws team diagrams from the code that already lists team members and what they can access. The diagrams stay up to date because they come from the same source as the real setup.
Read the CSV files that list the members of each team.
Read the YAML files that say which teams use which cloud accounts, and what they can do there.
Draw diagrams in layers, from the pods down to each person.
Redraw them automatically. One script makes new, dated SVG files whenever the data changes.
The examples below use made-up data.
Pods and teams
- Reactor Operations
- Mango: Core Systems
- Kiwi: Cooling
- Lime: Turbines
- Safety & Compliance
- Fig: Inspections
- Grape: Incident Response
- Corporate
- Plum: Finance
- Peach: HR
- Facilities
- Lemon: Maintenance
- Melon: Catering
- Research
- Cherry: Experiments
- Apple: Waste Disposal
Which pod owns which cloud accounts
| Pod | Accounts |
|---|---|
| Reactor Operations | reactor-monitoring-dev, reactor-monitoring-prod, cooling-systems-prod, turbine-telemetry-prod |
| Safety & Compliance | safety-alerts-dev, safety-alerts-prod, inspection-reports-prod |
| Corporate | payroll-prod, burns-manor-backup, donut-ordering-sandbox |
| Research | experiments-sandbox, waste-tracking-prod |
How I used AI
I used an AI tool (Amazon Kiro) as a coding partner. It:
- wrote the code that reads the files
- turned the data into GraphViz diagrams
- adjusted the layout when I showed it screenshots of broken results
- helped with tricky cases, such as people in more than one team
This took one afternoon instead of two to three days.
The AI did not do the thinking. I decided the structure, and I checked every result against the real organisation before sharing it.
Tools used
| Tool | What it does |
|---|---|
| Python | Reads the CSV and YAML files and builds the data |
| GraphViz | Draws the SVG diagrams |
| Terraform | Holds the team, account and access definitions |
| Amazon Kiro (AI) | Helped write the code and adjust the layout |
| Bash | Runs the script that redraws the diagrams |