Key result. Report turnaround went from 2 weeks to 48 hours, with 25+ reports delivered into daily production.
How it works
A data request arrives as a Jira ticket. An agent drafts the SQL and the report against governed data, an analyst reviews it, and it goes to production.
- Jira requestA stakeholder asks for data
- Agent draftSQL and report, using governed context
- Human reviewAn analyst checks and approves
- Production releaseA repeatable, documented workflow
It connects Claude to Snowflake, DataHub, Jira, GitHub and Figma through the Model Context Protocol (MCP) and Python.
What I built
- The workflow itself. The credit union's first agentic analytics workflow, from ticket intake through production release.
- Governed data context. Context in Snowflake and DataHub so agents can navigate the data and interpret metrics correctly.
- Human-in-the-loop review. Every report is checked and approved by an analyst before it ships.
Results
| Measure | Result |
|---|---|
| Report turnaround | 2 weeks → 48 hours |
| Reports in daily production | 25+ |
| Data requesters served | 500+ |
| Teams | Retail, branch and digital banking |
About this page
This is an overview of work done for my employer. It describes the workflow in general terms and includes no internal data, screenshots or system details.