I design AI systems that health plans can actually put into production.
20+ years building and architecting software, most recently for a U.S. healthcare payer. This site shows working AI applications for core plan workflows, with the evaluation, citations, human review and PHI handling a regulated environment needs.
One app for each core payer workflow
Prior Auth Copilot
Checks a prior authorization request against the coverage policy, one criterion at a time, and cites the evidence for a nurse reviewer.
LLM criteria review with verified citationsDenial & Appeals Assistant
Explains each denied line from its CARC/RARC codes, decides whether to appeal or correct it, and drafts the appeal from verified evidence.
Evidence-checked analysis and draftingBenefits Navigator
Answers coverage questions from the member's own plan documents, gets cost estimates from a calculator tool, and checks every citation and dollar figure.
Tool-using agent with checked answersPayment Integrity Radar
Finds providers whose billing stands out from risk-adjusted specialty peers, then writes a neutral, fact-checked brief for investigators.
Robust peer statistics with checked LLM briefsHCC Evidence Finder
Compares charts with submitted codes in both directions: missed conditions with MEAT evidence, and unsupported codes to review before a RADV audit.
MEAT evidence review with code-owned HCC rulesSame platform, same safeguards
Model output is checked in code
Every quote the model cites is matched against the source record. Routing rules live in code, not in the prompt.
People make coverage decisions
The AI recommends and routes. It never issues a denial; adverse determinations stay with clinicians.
Measured against labeled cases
Each app ships with a labeled test set and an eval script, so accuracy claims come with a number and a method.