Health Samurai Blog

How to configure and fine-tune a patient matching model in MDMbox — from defining match criteria and model structure to blocking, training, and manual tuning.

How Aidbox handles concurrent requests through its HTTP queue — three load scenarios, common failure modes, and how to tune worker capacity for stable performance.

FHIR's Patient/$merge assumes the server knows how to merge. Two decades of MPI vendor configs, EHR vendor divergence, and national registry policy show why one algorithm cannot serve every organization.

Give your coding agent a REPL inside its own runtime — and it will extend itself to fit you, your project, and the task.

FHIR R5's Patient/$merge is a start, but production MDM needs more. We built a resource-agnostic $merge with client-driven plans, atomic audit trails, and a generic $referencing operation.

We design agent tools as if agents were junior developers. They're not. Give them raw protocols and they'll outperform our carefully crafted abstractions.

Learn how we integrated WebMCP into Aidbox UI so AI agents can interact with the interface through declared tools instead of guessing from screenshots.

Store secrets in FHIR resources without storing the values. Aidbox resolves them from vault-mounted files at runtime. Secure, dynamic, and no vendor lock-in.

Why you need an independent FHIR server next to your EHR: analytics, sandboxing, granular sharing, and SMART on FHIR — without slowing the EHR.

In 2025, Aidbox focused on performance, FHIR conformance, and developer experience—removing bottlenecks and aligning deeply with the FHIR ecosystem. This post highlights the key architectural changes that make Aidbox faster, more transparent, and production-ready.