Bluesight's AI cut a 4,000-hour hospital compliance grind to 10-second queries

Hospital pharmacy compliance teams drown in paperwork. A single hospital subject to 340B drug pricing rules can burn over 4,000 staff hours annually just figuring out if they’re allowed to buy certain drugs through group purchasing organizations. It’s a mess of FDA shortage lists, inventory data, and backorder reports — all cross-referenced by hand.

Bluesight and AWS claim they’ve built a way out. Their new Prism Assistant, now live across 20 health systems, wraps AI around Bluesight’s existing ControlCheck software for monitoring controlled substances. Instead of assembling reports manually, compliance staff can just ask questions and get charts and summaries in return. AWS says Bluesight built the first version in a three-day sprint last September, with eight of its own engineers and seven from AWS. The result turned a process that could take five minutes of querying into something that takes about 10 seconds. “This is exactly what diversion program leaders have been waiting for—it gets them to answers faster and takes the manual grind out of every investigation,” said Bluesight’s Samir Neyazi.

The real technical story is what they didn’t do. Bluesight didn’t give a language model direct access to its databases. Instead, they wrapped existing APIs in AWS Lambda functions so the agent only sees structured data that’s already been processed by Bluesight’s own business logic. That architecture keeps the AI from hallucinating against raw pharmacy records while still making it genuinely useful for querying and reporting.

A bigger test is coming. A multi-agent system for 340B GPO compliance, which tackles that massive paperwork burden directly, is slated for 2026. In synthetic tests, AWS says it hit 100% invoice discovery and 93% evidence justification accuracy. Those aren’t production numbers, and nobody should mistake a clean lab test for what happens when real hospital data hits the system. But the fact that they built the whole thing in two days during a March engagement suggests the underlying pattern — connect agents to existing, well-structured APIs rather than raw data lakes — is becoming the default playbook for regulated industries.