AI for Medical Insurance

Decipher is an AI product that helps people actually understand their health insurance benefits — what's covered, what it costs, and where to go. Insurance answers, not clinical advice: the distinction mattered, and the product was built and positioned around it.

Decipher was one of the first AI chat systems for health insurance benefits. It raised $1.5M — a $1M round plus $500K from AI House — and signed one of the three largest insurance brokers in the United States. In a category where a wrong answer can cost someone real money or a needed prescription, the product had to be right, provably, every time.

Scott served as founding engineer on Decipher, with Dave as the second engineer on the project. Nullwest worked intensely on site, side by side with Decipher's CEO and COO, building the product end-to-end:

- A full RAG pipeline with custom chunking strategies, built to ingest and reason over 200–300 page insurance plan documents and drug formularies - A hybrid search and reasoning architecture that retrieves top matches across multiple documents, then runs a secondary pass to extract a conclusion — answering questions like "is this drug covered?" when the answer depends on step therapy and prior-authorization rules scattered across separate documents - A voice AI scheduling agent: a member describes a health issue, the system finds an in-network doctor on their plan, and the AI calls the office and books the appointment - A pipeline to ingest Transparency in Coverage files — sprawling multi-JSON datasets that had to be parsed and mined - Provider registry parsing tied to NPI numbers, plus third-party clearinghouse integrations that pre-verify a member's insurance and query deductibles and coverage limits - A multi-agent framework on AutoGen wrapped in a five-step safety harness to keep every response lucid and grounded

Scott also helped present and pitch, supporting the $1.5M raise. Voice AI, data engineering, compliance, and investor rooms — that range, from one team, is exactly what Nullwest was founded to offer.

A wrong answer about coverage is worse than no answer at all. Whether a drug is covered rarely lives in one place — it can depend on a formulary, a plan document, step therapy requirements, and prior-authorization rules, all in different files. We built a hybrid retrieval system that searches across multiple documents, then runs a secondary reasoning pass to extract a defensible conclusion instead of trusting the first match. The AutoGen multi-agent framework runs inside a five-step safety harness, because in this category, confident-and-wrong is the failure mode that kills trust.

Healthcare data doesn't arrive clean — or small. Plan documents run 200–300 pages. Transparency in Coverage files are enormous multi-JSON dumps. Provider registries and eligibility systems each speak their own dialect. We built custom chunking for the documents, dedicated parsing pipelines for the federal data, NPI-linked provider records, and clearinghouse integrations that pre-verify insurance before a member ever asks a question — so the chat experience feels simple because the data work underneath it isn't.

Put "AI" and "health" in one sentence and people hear "diagnosis." The fastest way to lose users and regulators is ambiguity about what the product is. Decipher was deliberately built and positioned as an insurance benefits system, not a clinical one — and shipped HIPAA compliant and SOC 2 compliant, so the trust story was an engineering fact, not a marketing claim.

Decipher raised $1.5M — a $1M round plus $500K from AI House — with Scott in the room helping present and pitch. It signed one of the three largest insurance brokers in the United States (the name is under NDA, but there are only three candidates). And it shipped as one of the first AI chat systems for health insurance benefits, HIPAA and SOC 2 compliant from the start. [Optional: usage metrics or client quote here.]

Building an AI product on regulated, messy, high-stakes data? The team that built Decipher's pipelines, agents, and safety harness from the inside is the same team that has shipped for Spotify, Vice, Intel, and Xembly. Let's talk about yours.