Published daily
![The Deployment Brief]()
The daily brief on AI that ships. The news that matters if you build these systems, then one recent paper explained from the ground up.
00:00 / --:--
Meridian Applied is a physician-led practice that deploys AI, quantitative, and data systems into high-stakes environments. We build working systems, tested against reality, not decks.
Most AI never leaves the demo. The gap is rarely the model. It's the data plumbing, the evaluation, and the judgment about where AI belongs and where it doesn't. That gap is the whole job.
Retrieval systems over your own material, agent workflows, and fine-tuned models, built with the evaluation harnesses that prove they're reliable before anyone leans on them.
Financial modeling, statistical analysis, and decision-support built on real data. From optimization models to backtested strategies to dashboards leadership actually reads.
Designing how the pieces connect: data standards, integration between platforms that were never built to talk, and roadmaps that survive contact with a real organization.
For environments where sensitive data can't leave the building, we deploy locally hosted open-weight models that keep your data yours, without giving up capability.
The measurable time saved is rarely the point, and it's often small. The real value is what a good system takes off someone's plate: the scattered spreadsheet, the shifting rule nobody's tracking, the detail they're afraid to forget. A clinical scribe doesn't save a physician much time either. What it removes is the burden of holding the whole visit in memory. That's the standard every build here is held to.
The first version is built for one organization, against their actual systems, until it produces something they'd stake a decision on. Patterns get reusable only after they've earned it.
A system that's right most of the time is a liability where the stakes are real. Every build ships with a way to measure whether it's actually working, expressed as a number.
The practice was forged in healthcare and humanitarian operations. Sensitive material is handled with the judgment that background demands, and stays where it belongs.
The daily brief on AI that ships. The news that matters if you build these systems, then one recent paper explained from the ground up.
Identifying where health data fragments across systems and agencies, then designing the architecture and buy-in to close it. Time at the UN carried this to roadmap-level approval, and surfaced a gap that turned out to be universal, now the throughline into ongoing policy research.
Retrieval systems that let front-office staff ask plain-language questions of their own scouting and research material and get trustworthy, sourced answers on the clock.
Serving as AI implementation specialist for a university civil and environmental engineering department, bringing practical generative AI into research and operational workflows.
Portfolio optimization and backtesting frameworks, and analytical tooling that turns messy inputs into decisions leadership can defend.
A retrieval system combining dense vector search with a SQL-based concept graph as an independent path, testing where the two approaches complement each other and where they diverge.
Try it live →Proposes a cooperative governance model for health data exchange, modeled on SWIFT financial-messaging infrastructure.
Read on SSRN →Identifies where patient-data continuity breaks between field triage and hospital intake, and proposes an NFC-based offline tracker grounded in humanitarian operations.
Read on SSRN →Meridian Applied is principal-led. The work is done by someone who sits at an unusual crossroads: a physician who builds AI systems himself, in code, and has deployed them where a wrong answer has real consequences.
That combination is the point. Clinical judgment about what "good enough" means when stakes are high. The engineering to build the thing rather than specify it. And the executive fluency to sit across from a chief medical officer, a general manager, or a head coach and translate an ambiguous problem into something that ships.
Specialist collaborators are brought in when a project calls for it. The accountability stays in one place.
The best engagements start with a specific, stubborn problem. Bring yours and let's see if there's a system in it.
Get in touch