Heath Reinhard
Documentation Engineer | AI Knowledge Systems | Eval-Driven Content Analytics & Optimization
I build eval-driven knowledge systems that improve production AI-agent answers by improving the content they retrieve.
At Meta, my recent work focused on a simple idea: when an AI agent gets an answer wrong, the fastest path to a better answer often runs through the content, not the model.
Examples of recent work
- AI knowledge optimization system: evaluated production AI agent trajectories, identified knowledge gaps, generated source-grounded fixes, routed changes through review, and re-evaluated outcomes across 49 internal products.
- AI documentation consumption dashboard: measured how internal AI assistants retrieved and cited documentation across 300M+ retrievals, 54M+ documents, and 19 document types.
- Documentation health analytics platform: increased coverage from 44.1% to 73.9% across 600+ internal products.
- Claude Code plugin suite: shipped 10 skills for RAG evaluation, gap analysis, remediation workflows, semantic clustering, golden-dataset generation, and production AI quality measurement.
AI outcomes depend on documentation quality
AI answer quality often depends on the quality of the knowledge it retrieves. My work sits at that intersection: evaluation, retrieval quality, documentation analytics, and content optimization.
I came up through writing and editing, which I think is an underrated background for AI evaluation work. The people who notice when an answer is almost right are often the same people who notice when a sentence is almost clear.
I’m equally comfortable in a SQL query, a CLI coding agent, or a draft that needs to be cut by 30%.
Background
Before Meta, I translated computing history for audiences ranging from kindergarteners to professional engineers at Living Computers: Museum + Labs in Seattle. Before that, I ran an IT services company for nine years.
Explore
If you think I might be the right fit for your organization, get in touch.
About this site
This site was created with Markdown, Git and GitHub, Jekyll, and Claude Code using a docs-as-code workflow.