A governed imaging-to-edge workflow.
TrichoEdge is a local-first engineering system spanning guided multi-illumination capture, an on-premises inference appliance, a governed model pipeline, and privacy-minimized report delivery.
A governed local imaging workflow from guided capture to edge inference.
Context
The system, not just the model.
The project explores how a specialized imaging workflow can remain local while preserving contracts, expert oversight, data governance, hardware control, and release evidence across several independent components.
System surface
What the work covers
- Native guided-capture terminals and private offline queues
- A customer-site appliance for authoritative storage and bounded inference
- Consent-aware data, expert annotation, training, and evaluation tooling
- Compact multimodal Qwen adaptation on DGX-class infrastructure
- Signed packages, hardened services, health checks, and rollback
- A static phone viewer that receives minimized signed data locally in the browser
Engineering lessons
What it demonstrates
- The input and output contract must be shared across capture, training, runtime, and reports.
- Synthetic bring-up and product evidence need unmistakably different states.
- Model selection, accelerator conversion, hardware validation, and release are separate gates.
- A privacy pattern is incomplete until the viewer origin and update path are in the threat model.
Public claim boundary
What we are—and are not—saying.
Current operation is explicitly synthetic and engineering-only. TrichoEdge is non-diagnostic, and no clinical efficacy, validation, or production-readiness claim is made.
Current operation is synthetic and engineering-only. Product model promotion, production trust, and real-client use remain closed behind separate prerequisites.
Start with the constraint
Need this kind of joined-up local engineering?
We work across the device, data, model, application, and release boundary—while keeping maturity and risk visible.