Two dashboards: AI operations and government spending

I needed two dashboards for two very different audiences. One tracks the health of AI agents and services for technical operators. The other tracks government spending across state agencies for the public.

The platform data was scattered across Langfuse, Portkey and internal run history. The government data sat in PDF documents, public records and agency databases.

What I built

For operations: summary cards for agent cost, run count, success rate and latency; breakdowns by agent, model and time period; pipeline analytics; and a filtered log viewer.

For government data: Python pipelines that use PyMuPDF to parse PDF documents into structured data, and pandas, matplotlib, plotly and seaborn for analysis. The data model has three layers: people (headcounts and titles), infrastructure (leases) and contracts. The dashboard shows spending trends, agency comparisons and headcount changes. It is open source and publicly accessible.

What came of it

The operations dashboard gives one view of agent costs, performance and health. The government dashboard is a public transparency tool that tracks state spending across more than 6 agency categories. Both serve different audiences from the same technology patterns, in TypeScript for the live views and Python for extraction and statistics.