Agent Almanac
A searchable catalog of 9,177 MCP servers for AI agents — ingestion pipeline, full-text search, verification checks, deployed and serving. Built end-to-end: schema, crawler, dedupe, site.
13+
years in financial & healthcare data
Senior healthcare-data analyst — claims, encounter files, enterprise reporting — who also builds production AI systems that actually run: knowledge graphs, retrieval pipelines, autonomous agents. I make messy organizational data legible, then automate the reports nobody should be doing by hand.
nodes = where I've worked · edges = what carried over · leaves = things you can inspect
Not slideware. These run today — some public, some on infrastructure I operate.
A searchable catalog of 9,177 MCP servers for AI agents — ingestion pipeline, full-text search, verification checks, deployed and serving. Built end-to-end: schema, crawler, dedupe, site.
101K agent profiles, 1M posts, 880K comments organized into 6,700 communities, with a daily automated sync pipeline and a custom MCP server exposing semantic search over the whole thing.
A self-operating agent with 950+ logged sessions, 90+ custom tools, graph-backed memory, token-budget management, and health monitoring. Runs 14 scheduled sessions a day unattended.
Full patient pipeline for a medical imaging practice — booking, automations, ten working views, staff training — delivered 32 hours from kickoff.
Content-analysis platform: paste a URL, get a structured breakdown of what makes it work — with the methodology published, because scores without explanations are noise.
The nine-to-five that built the rigor: regulated data, real deadlines, teams.
For organizations that need the thing done, not a seat filled. Everything scoped in writing before any money moves.
Data cleanup, consolidation, and a working dashboard (Power BI or Looker Studio) your board or funders can actually read — plus the recurring reports automated so staff stop rebuilding them monthly. Nonprofits: Power BI is already free to you through the Microsoft grant; setup is the missing piece.
837P / 999 / 277CA file validation, rejection triage, reconciliation between what you sent and what got paid. Thirteen years inside a national payer's reporting function — I've seen how these files break.
Send a paragraph about the problem — the report that eats a week, the files that won't reconcile, the dashboard nobody trusts. You'll get a straight written answer: what I'd do, what it costs, or a pointer to someone better suited if that's the truth.
✓ async by design — clear writing beats meetings