Back to selected work

PROJECT BRIEF / 04

Evidence behind the extraction

Source-grounded PDF chart intelligence

Prototype2026Public repository
  • Azure AI Search
  • GPT-4o
  • Custom skills

The context

Extracting a value is not enough when a reviewer needs to know where it came from. This prototype focuses on source-grounded interpretation of synthetic financial charts.

The approach

  • Use Azure AI Search indexing and a custom Web API skill.
  • Analyze normalized chart images with an existing GPT-4o deployment.
  • Return structured results including visible values, source pages, evidence, confidence, and review flags.
  • Keep regional deployment requirements and infrastructure prerequisites explicit.

The leadership connection

An evidence-first design for AI-assisted document understanding, with reviewability and deployment constraints treated as part of the architecture.

What to look at

Inspect the design

Review the published architecture and structured extraction output.

Separate the test paths

The local image preview and the PDF-to-Search indexing workflow validate different parts of the system.

Check current prerequisites

Confirm regional access, identity, hosting quota, and deployment status in the repository before running.

Continue exploring

Useful AI. Deliberate controls.

Next project: Consent-aware campaign application