singhankit491/otovision-mlops

OtoVision MLOps — End-to-End Medical Imaging Deployment

Author: Ankit Kumar Singh Positioning: Computer Vision • Medical AI • MLOps • Docker • Kubernetes • Responsible AI • Deployment Engineering

OtoVision is an end-to-end otoscopic medical-imaging MLOps portfolio project. The public Hugging Face Space is a free client-side workbench that performs image-quality inspection directly in the browser and documents the model-readiness, evaluation, API, container and orchestration path implemented in GitHub.

Evidence boundary: A genuine five-class trained checkpoint is not bundled in the public Space. The live demo therefore does not fabricate disease predictions or publish unsupported clinical metrics.

End-to-end flow

Permitted otoscopic image
→ browser-side image QA
→ deterministic preprocessing contract
→ model-readiness / checkpoint provenance
→ genuine checkpoint inference when available
→ saved predictions + evaluation
→ FastAPI
→ Docker
→ Kubernetes
→ GitHub Actions release validation
→ Hugging Face portfolio Space

Current evidence

The repository is designed to export balanced accuracy, macro/weighted precision-recall-F1, multiclass ROC-AUC, sensitivity/specificity, confusion matrix, ECE, Brier score and selective-coverage outputs from a genuine run.

Responsible use

This is a research/engineering portfolio demonstration, not an autonomous diagnostic system. Clinical use would require approved datasets, patient-level leakage controls, external validation, clinical oversight, security/governance review, monitoring and applicable regulatory approval.

• —GitHub: https://github.com/singhankitsrf/Otovision-MLOps
• —Space: https://huggingface.co/spaces/singhankit491/otovision-mlops