GuideAI Health — Vascular AI for an active FDA submission
Principal ML engineering embedded in the sponsor's program: TensorHarmony owns the research-to-production ML pipeline for GuideAI Health's first FDA submission, with a clinical-review layer for CTA run-off studies—carried out inside the sponsor's FDA-aware production environment.
Status: active engagement
What follows reflects scope and responsibilities at the level disclosed on the principal's CV and LinkedIn. Proprietary architecture, model performance, dataset composition, validation metrics, and other GuideAI Health intellectual property are intentionally not published here and remain confidential. Systems referenced are owned by GuideAI Health.
The challenge
GuideAI Health is a US clinical-AI company focused on AI-driven diagnosis of peripheral artery disease and arterial occlusion. When TensorHarmony came in, the company had early AI prototypes that had proven the underlying clinical value, but were not yet structured as a unified, production-ready ML platform — and were not yet aligned to the rigour required for a first FDA submission.
GuideAI needed a senior, hands-on ML lead who could bridge applied research and production engineering, lead the technical case for the regulatory submission, own the path from prototypes to deployable software, and build the clinical-review tooling needed for internal and investor-facing reviews.
Scope of the engagement
Principal ML engineering, embedded full-time: Omid Sakhi (via TensorHarmony) operates as the company's first full-time technical hire and principal ML engineer.
Applied ML lead for arterial occlusion detection — model development, retraining, evaluation frameworks, and ensemble methods.
FDA-submission ownership for the AI component: alignment between model design, validation methodology, and regulatory expectations for the indication.
Research-to-production pipeline: training, retraining, evaluation, packaging, and the path to deployable releases — working closely with production engineering.
Clinical-review and demo applications used by leadership and investors to walk through real cases end-to-end.
What was built (selected, public-disclosure level)
The engagement spans three publicly described tracks—stopping at the level appropriate for an active program under confidentiality. All software and IP described here are owned by GuideAI Health; TensorHarmony is the engineering counterparty.
Vascular research pipeline
A pluggable Python pipeline for medical image analysis on vascular CTA — DICOM ingest, vessel cleanup, landmark detection, occlusion detection, ensemble decisioning, and result export. Containerized for AWS Batch with S3-backed I/O.
Production pipeline
A modular Python package layout for the production pipeline — core engine plus per-node packages (DICOM loaders, segmentation, landmark assignment, vessel statistics, occlusion ensembling, exporters) published as private packages with CI-driven release on each change.
VascularAssist clinical-review demo
A production-style web app for reviewing CTA run-off studies: 3D + axial viewing, PDF reports, DICOM upload, async processing pipeline, optional AWS Cognito auth, ECS / CloudFront / Terraform deployment.
Associated IP
CONFIDENTIAL
Provisional patent application · Filed 2026 · Inventor
Why this engagement matters for clients
This is what current FDA-aware ML engineering looks like in practice — not in retrospect. The work happens against live regulatory expectations, real validation timelines, and a production pipeline that has to survive review.
For TensorHarmony clients, that means the strategy, hands-on, and review modes are informed by an active submission, not by a program that ended years ago. The frameworks, the documentation discipline, and the trade-offs are current.
It also shows the studio working at the boundary between research-quality ML and shippable software, with a clinical-review demo built in the same engagement. That sequencing — research, pipeline, then a real reviewable interface — is the same pattern we bring to non-medical AI products.
Discuss Your Next System or Deployment
Technical discovery, production engineering, and deployment-focused collaboration for intelligent systems, computer vision, and medical imaging workflows.
No obligation — just a focused technical conversation