ML Health AI designs, develops, and delivers software tailored to each healthcare client's precise requirements, compliance posture, and operational reality — from revenue cycle and document automation to lab automation and hands-free AR capture.
MarginLogic is built by a team of healthcare operators, ML engineers, and HL7 integration specialists. We ship one thing: automated intake that your compliance officer trusts.
A purpose-built workspace for clinical paperwork. Extraction, reconciliation, routing, and audit — all in one pipeline your team can actually audit.
Explore services →Reference labs, multi-site clinics, health systems, and payers — anywhere requisitions, referrals, or claims forms need to move from fax to system without a human re-keying them.
See use cases →Founded in 2026 and headquartered in Dallas, TX. Our engineering team is primarily based in the United States, with partner teams worldwide so we can deliver and support customers on a 24-hour cycle.
About us →Fair questions. Here are the honest answers — the same ones we give in the first sales call.
Off-the-shelf OCR is great at reading clean documents. It's terrible at the four things clinical intake actually requires: handwritten requisitions, EMR reconciliation, rules-based routing to specific lab queues, and an immutable audit trail your compliance officer can defend.
We've evaluated every major OCR API against real clinical documents from our pilot customers. Accuracy on handwritten reqs ranges from 71% to 88%. Ours is 97.6% — and it comes with the rest of the pipeline attached. "Off the shelf" gets you a 15-minute demo. It doesn't get you a production deployment.
Yes. We develop under an ISO 13485:2016-aligned quality management system. Our QMS covers design controls, risk management (ISO 14971), change control, supplier qualification, CAPA, and post-market surveillance. We hold documented evidence of each process and make our QMS manual available under NDA.
For customers who need a formally certified supplier, we're tracking a Q3 2026 notified-body audit. Customers building medical devices that incorporate our software receive technical files suitable for their own 13485 submissions.
Yes. Our software lifecycle process is built against IEC 62304. Components are classified per Class A/B/C depending on their role in the pipeline — extraction and reconciliation modules are Class B; routing and audit modules are Class C because an error could contribute to patient harm.
We maintain the full 62304 deliverable set: software development plan, requirements, architecture, unit & integration verification, risk traceability matrix, SOUP inventory, and release documentation. Customers submitting regulatory filings receive these artifacts as part of their technical file package.
In the product: AI is the engine behind extraction, reconciliation, and classification. Every prediction carries a confidence score. Anything below a configurable threshold is flagged for human review — so a clinician or operator always makes the final call on ambiguous cases. We never auto-approve a clinically-significant decision without a deterministic verification step behind it.
In our own development: We use AI assistants for code generation, documentation, and test synthesis under strict guardrails — all generated code is reviewed, all generated tests are validated against known-good fixtures, and no PHI is ever routed through a third-party model. Our 62304-qualified release pipeline treats AI-authored code the same as human-authored code: it has to pass the same verification gates.
In delivery: AI never replaces a deployment engineer. Every new customer goes live with a forward-deployed human on our team who has operated the pipeline in staging before it touches production.
Bring 500 sample requisitions. We'll extract, verify, and route them against your EMR in a sandbox — so you can see the numbers before you sign anything.