Technology Financial Management
Multi-Agent System
I built a coordinated set of custom AI agents to support financial analysis, portfolio reporting, and evidence review. A central agent routes requests to specialist roles and brings their outputs together for team review.
Explore the system +
The problem
Financial questions often span capital spending, operating expenses, business value, and regulatory evidence. Reviewing these areas separately makes it harder to keep assumptions, sources, and follow-up work aligned.
My contribution
I built the TFM agent system using custom agent definitions, specialist instructions, shared routing rules, and structured work and evidence registers in VS Code.
Five coordinated roles
- TFM Orchestrator
- Identifies the request, delegates work, tracks follow-ups, and reconciles specialist outputs into a combined answer with sources and identified data gaps.
- Capital Specialist
- Supports capital actuals and forecasts, project-health reviews, variance analysis, prioritization, and reporting consistency.
- O&M Specialist
- Supports operating-expense reconciliation, forecast alignment, variance analysis, risk and opportunity tracking, and recurring reporting quality checks.
- Rate Case Specialist
- Supports evidence collection, filing reconciliation, gap identification, and draft reporting materials for human review.
- VRO Liaison
- Connects cost analysis with benefit evidence from a separately owned value-realization assistant. Distinguishes realized benefits, forecasts, cost avoidance, and opportunities.
How it helps the team
Provides a single entry point for related financial questions, supports repeatable analysis, keeps assumptions and sources traceable, and makes pending work and data gaps easier to follow. Outputs support analyst and leadership review.
Delivered capabilities
- Coordinated operating model: One orchestrator and four specialist interfaces, supported by process documentation and registers for assignments, working files, and evidence.
- Document indexing and validation: A capital whitepaper register with reconciliation checks, file scanning, and review flags for uncertain classifications.
- Financial reporting: Portfolio reporting that brings together expenses, capital, project status, benefit evidence, forecast accuracy, risks, and data-quality checks.
- Evidence-based planning: A progress dashboard that tracks commitments against recorded evidence while preserving the official planning board.
- Review readiness: Reporting and evidence checks that surface missing explanations, inconsistencies, and outstanding actions before handoff.
Rollout and review status
The agent system is implemented and in use. Document registers and reporting outputs go through owner review. The Power BI report is built, with service rollout still in progress.
Supporting tools
Alongside VS Code custom agents and GitHub Copilot, the documented workflows use Python, PowerShell, Excel automation, HTML/CSS, JavaScript, Power BI, and structured JSON/CSV files for analysis and reporting.
Review and controls
The documented workflow includes source citations, data-quality checks, specialist boundaries, and human review. The O&M role proposes changes without editing source workbooks; regulatory materials remain drafts for appropriate review.