Hammer Labs
Hammer Labs provides a verification platform that tests healthcare AI agents by identifying when they produce incorrect outputs and when they should refuse to answer. The platform builds answer keys from preserved payer evidence—including billing codes, median rates, and fee schedules—so payers can measure AI performance against real-world claim data and meet CMS audit and decision-deadline requirements. Its hosted world operates as an MCP server that returns figures with receipts showing row counts and data vintages.
- Insurance Technology
- Healthcare Technology
- Data & Analytics
- Regulatory & Compliance Technology
- Artificial Intelligence
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Healthcare payers are deploying AI agents to automate coverage and pricing decisions, but these agents frequently produce confident-sounding answers that are wrong or unsupported by evidence. Real claims rarely carry a clean answer key, so organizations still pay nurses, coders, or medical directors to check every AI output, creating an unsustainable review burden. New CMS regulations under CMS-0057-F impose decision deadlines and audit requirements that make measurable AI verification a compliance necessity rather than an option.
Solution
Hammer Labs provides a verification platform that tests when healthcare AI agents are wrong, when they should refuse to answer, and where human review still belongs. The platform builds an answer key from preserved payer evidence—capturing each source with its date, basis, and original artifact before upstream files are replaced—then runs agents against records with known answers and failure conditions. It returns figures with receipts attached, showing row counts, vintages, and the exact evidence behind each answer, and refuses by name when a question falls outside what its sources can settle. The system is delivered as an MCP server that connects directly to AI agents, enabling them to check claims against the evidence corpus in real time. Public evidence demonstrates the method with 115 billing codes and 9 payers, while private evidence runs the actual production work.
Target Audience
Primary customers are healthcare payers, including health insurance companies and managed care organizations, that deploy AI agents for coverage determinations, prior authorization, and pricing decisions and need to meet CMS audit and decision-deadline requirements.
Features
- MCP server integration that connects to AI agents via a plugin, with no corpus download or API key required
- Receipt system that attaches row counts, data vintages, and source artifacts to every figure returned
- Refusal engine that returns "unknown" when evidence is missing and names what would settle the question instead of laundering absence into confidence
- Rate tools covering 115 billing codes across 9 payers including Aetna, Anthem, Cigna, and UnitedHealthcare, positioned against Medicare fee schedules
- Hospital filing tools for standard charges under 45 CFR 180, including filer inventory and variance analysis
- Evidence tools for hypertension, prognostic associations, and treatment effects that return one published effect size at a time with interval, population, and PMID
- Calculators over declared inputs for screening yield, deferred versus avoided costs, regression to mean, competing event coupling, measure reliability, and withhold earnback
- Vintage tracking that preserves the exact fee schedule and rate versions that existed at the time of each query