Ontario Infrastructure Decision Intelligence Platform
Making public infrastructure data useful for decision-making
The problem
The problem
Ontario publishes its infrastructure project disclosures publicly, but the raw data is hard to actually use: budgets are missing on a large share of records, completion dates are often imprecise, and there's no structured way to ask real questions of it, like which project types are furthest behind or where funding information is simply absent.
The approach
An automated pipeline ingests disclosures from Ontario's public CKAN data API, processes them in Python, and loads them into a five-table dimensional model in PostgreSQL. A FastAPI service exposes read-only analytical endpoints over that model, and a React/TypeScript frontend with an interactive Leaflet map lets someone explore projects by location, type and status rather than scrolling a spreadsheet.
Implementation
Implemented
- Automated ingestion from Ontario's public CKAN API
- Five-table PostgreSQL dimensional schema
- Eight automated data-quality checks run against every ingest
- FastAPI analytical endpoints
- Interactive React/TypeScript + Leaflet map and dashboard
Engineering decisions
Engineering decisions
Missing data stays visible, not hidden
About 47% of projects report a literal $0 budget in the source data. Rather than treating that as a real value or silently dropping it, it's flagged with an explicit budget_undisclosed indicator so analysis can't mistake "not reported" for "free."
Imprecise dates modeled honestly
Roughly 96% of completion dates in the source lack day-level precision. Instead of forcing them into a plain DATE column (which would fabricate a precision the data doesn't have), each date carries a precision field alongside an upper-bound value.
Validation catches real inconsistencies
Automated checks caught at least one literally impossible date in the source disclosures (June 31st doesn't exist) — the kind of record that breaks naive date parsing silently rather than loudly.
Results and status
Results and status
This is a self-initiated engineering project, not a client engagement, so there's no business outcome to report. What's verified directly from the repository: a working end-to-end pipeline against Ontario's public dataset, eight automated data-quality checks, and a documented audit of the data's real limitations rather than a polished but inaccurate picture of it.
Contributors & attribution
Contributors and attribution
The original 2024 analysis was a group project credited to Tobi Oladimeji, Sushma Gouri Deivanayagam, Sachin Suresh, Dwij Shyamsunder and Surya Amit Malwade. The 2026 rebuild — the version this case study describes — is Tobi's independent work, built using AI-assisted engineering tooling, on top of the data model and findings from that original project.
Have a similar operational challenge?
Tell us what's happening and we'll tell you honestly whether it's something we can help with.