In the high-stakes environment of hospital pharmacy management, compliance audits have long been a grueling, manual endeavor. Pharmacists and compliance officers routinely spend over 4,000 hours annually cross-referencing fragmented data sources—ranging from FDA shortage lists and American Society of Health-System Pharmacists (ASHP) data to internal inventory levels and machine learning-based shortage predictions across 620 hospital networks. For Bluesight, a leader in medication management, this manual bottleneck was not just an operational inefficiency; it was a barrier to real-time safety and oversight.

Scaling Compliance with Prism Assistant

To modernize this workflow, Bluesight developed Prism, an AI-driven layer designed to integrate and reason across six distinct healthcare product datasets. In May 2026, the company launched the Prism Assistant for ControlCheck, a product that monitors controlled substance transactions within hospital pharmacies to detect potential diversion patterns. Previously, compliance teams had to manually correlate dashboard signals to generate reports. With the new AI assistant, that process is condensed into a conversational interface that delivers results in seconds.

Currently deployed across 20 healthcare systems, the solution bridges the gap between external regulatory data and internal procurement logs. By shifting from manual spreadsheet reconciliation to an agentic workflow, practitioners can now query the system directly, receiving verified insights backed by real-time data integration rather than static, retrospective reports.

Orchestrating Multi-Agent Workflows with MCP

To maintain high accuracy, Bluesight moved away from a monolithic model approach, which often suffers from knowledge dilution in specialized domains. Instead, they implemented the Amazon Bedrock AgentCore framework. This architecture utilizes a Coordinating Agent that parses user intent and delegates specific tasks to specialized agents, such as CostCheck, ShortageCheck, and 340BCheck. By isolating these domains, Bluesight significantly reduced the risk of model hallucinations.

To ensure interoperability, the company used the AgentCore Gateway to convert existing product APIs into tools compatible with the Model Context Protocol (MCP). This allows agents to discover and invoke tools autonomously with the correct parameters, drastically reducing the engineering overhead required to integrate new product lines. The runtime is hosted on a serverless infrastructure, featuring session isolation that ensures data from multiple hospital systems remains strictly segregated, even during concurrent queries.

Optimizing Performance via API Wrapping

One of the most critical technical shifts was the move from direct database access to an API-wrapped architecture using AWS Lambda. By funneling data through Lambda endpoints, Bluesight reduced query latency from 5 minutes to just 10 seconds. This design choice serves a dual purpose: it prevents the AI from generating inefficient or hallucinated database queries, and it keeps business logic firmly within the application layer, ensuring that the AI only interacts with validated, structured data.

This rapid development was facilitated by the AWS Experience-Based Acceleration (EBA) program. A joint team of 8 Bluesight engineers and 7 AWS experts completed an intensive 3-day sprint to establish the initial architecture. By utilizing Infrastructure as Code (IaC) and robust observability tools, the team reduced the typical AI product development lifecycle from 12–18 months to under 9 months.

Security and Deterministic Design

For Bluesight, HIPAA eligibility was the non-negotiable baseline for adoption. The company signed a Business Associate Agreement (BAA) with AWS to ensure legal compliance and configured Amazon Bedrock to explicitly exclude customer data from foundation model training. This ensures that sensitive hospital information is never leaked or incorporated into responses for other clients. Furthermore, the deployment resides within a Virtual Private Cloud (VPC), providing physical network isolation and end-to-end encryption.

Beyond security, Bluesight prioritized a "deterministic AI" design. In this model, the AI acts as a conduit for verified data rather than an autonomous decision-maker. By keeping business logic in the application layer and using agents strictly for orchestration and tool invocation, the system provides an audit trail for every inference. This predictability is essential in regulated industries where the cost of an error is measured in patient safety and legal liability.

In highly regulated sectors, the success of an AI agent depends on the integrity of its data access path. By prioritizing deterministic system design over raw model performance, organizations can build AI tools that are not only fast but fundamentally reliable.