The company story
A Canadian government housing agency that depends on documents — lots of them, across siloed repositories. The work was to turn that archive into something staff could query and trust — starting with Project Approval Documents.
PAD classification confidence (POC)
≥95%
Classification
≥95% conf.
Delivery
2-phase POC
Search
RAG + metadata
A Canadian Government Agency — impact snapshot
The pressure
What was on the line
Manual extraction and scattered repositories created delays and error risk. Quarterly reporting cycles absorbed weeks of hand work. Housing operations need reliable process — not a flashy pilot that falls apart when the data gets messy.
What changed
The arc of the work
- 01
Make the documents searchable
Custom classification models and a RAG pipeline with semantic search turned siloed repositories into a knowledge base staff could actually use.
- 02
Prove classification confidence on known formats
Azure Document Intelligence plus tailored classification reached ≥95% confidence on known Project Approval Document versions in the POC — enough to trust the retrieval path.
- 03
Design for reporting cycles that used to take weeks
The POC path was built to shrink ministry reporting cycles that previously meant weeks of manual extraction — with a documented production route (Canada East, hybrid search, version-specific models).
Before and after
What people could feel in the week
Siloed repositories and weeks of manual extract for reporting
→Searchable archives with classification + RAG on known PAD formats
Staff hunting documents by hand
→Query the agency’s own corpus with metadata filters
Where they are now
The outcome that stuck
The POC proved the pipeline: classify, retrieve, and answer from the agency’s own documents. The production path is documented; the demo is not the deliverable.
For technical readers
How it was built
Azure Document Intelligence, custom classification models, RAG with semantic search and metadata filters over agency document repositories. POC delivery via New Value Solutions.
Azure Document Intelligence · Custom classification · RAG / semantic search · Azure AI Search
More company stories
- Connected TutorsA growing tutoring company got capacity back — without turning owners into a full-time ops desk.
- CopperTeamsCopperTeams’ 20% team-performance guarantee — extended with AI coaching between human sessions.
- Athyna InsightExpert-validated building code answers in seconds — with licensed professionals in the loop.