Index

Applied AI / Frontend

Year
2026
Built
DeepLabV3+ · API · UI
Context
Five-person research
Status
Non-diagnostic prototype
Designed orange MIRO cover reading AI medical imaging assistant above a standing person and a subtle cross-grid texture.

Overview

A research interface built around DeepLabV3+.

MIRO began as a five-person master's research project. The work shown here combines the DeepLabV3+ binary and multiclass branch, FastAPI service, Next.js interface, and deployment I built. Teammates developed the additional models and broader academic pipeline.

Built with

  • Next.js 16
  • TypeScript
  • Tailwind CSS 4
  • GSAP
  • Three.js
  • FastAPI
  • PyTorch
  • DeepLabV3+

MIRO product tour

A tour of MIRO, from its public experience and anatomy selection to the image-upload workspace.

System

One traceable path from experiment to review interface.

This map covers the DeepLabV3+ branch I implemented: experiments, inference, frontend, and deployment.

  1. 01

    DeepLabV3+ experiments

    Binary and T1–L5 multiclass training and evaluation notebooks.

  2. 02

    Standalone FastAPI inference

    Six polygon/raw endpoint modes expose the presented branch.

  3. 03

    Server-side Next.js proxy

    Protected requests connect the product without exposing backend details in browser code.

  4. 04

    Interactive review workspace

    Overlays, layer controls, zoom, pan, and technical output views.

01 / Product problem

A clear start to the research flow.

Anatomy selection and input preparation establish what the prototype can do before an image reaches inference.

  • Anatomy and input selection
  • Non-diagnostic context before inference
The opening decision narrows the research workflow to the available spine-segmentation module before an image enters the workspace.

Input workspace

A neutral workspace before inference.

The interface isolates input preparation from model output, keeping the next available action clear without implying a result.

A reviewed local demonstration input enters the protected workspace before an output mode is selected.

02 / Model to interface

From inference output to useful comparisons.

Six polygon and raw modes turn DeepLabV3+ responses into results that can be compared directly.

  • Binary and T1–L5 multiclass modes
  • Raw-mask and polygon representations
The setup panel maps one input to binary or multiclass output and keeps raw and polygon review modes explicit.

Result comparison

The same output stays directly comparable.

Polygon overlays improve legibility while the raw prediction remains available as a separate technical representation.

One matched result can be reviewed as structured polygon overlays or switched to its raw prediction representation.

03 / Technical review

Keep raw model output available for inspection.

This view exposes the prediction directly for technical inspection without framing it as a clinical conclusion.

  • Direct pixel-level model output
  • Explicit non-diagnostic context
The raw view preserves direct access to model output alongside the more readable polygon representation.

Beyond the inference workspace.

Public pages, navigation states, and additional review views complete the broader research experience.

Drag to scroll