Emmanual Antony Clement

A phone enters the frame with its screen off. Phone Visible reaches 100%, a low-severity phone flag opens alongside the resolved gaze flag, and the risk score settles at 54 (Elevated).

1 / 5

A privacy-first, client-side AI proctoring engine where all computer vision runs locally in the browser—no continuous video stream ever leaves the device, and only structured risk telemetry and still evidence snapshots reach the proctor console. The system orchestrates three isolated Web Workers running on independent throttles so the UI stays locked at 60 FPS: YOLO26 (ONNX Runtime + WebGPU) for object and multi-person detection, MediaPipe FaceLandmarker for 3D head-pose and iris gaze tracking, and a candidate-masked frame-differencing worker for foreign-object and background anomaly detection. Built on a zero-copy video pipeline (MediaStreamTrackProcessor and Transferable VideoFrames), it feeds a six-signal hysteresis risk engine that accumulates temporal evidence and filters transient noise into an explainable live dashboard and exportable audit trail.

ONNX Runtime · WebGPU · YOLO26 · MediaPipe · JavaScript · Web Workers

Roastery overview with live totals for green beans, roasted batches and low-stock alerts, plus recent transfers and critical alerts

1 / 4

Full-stack inventory management for a working coffee roastery, live with production data. Tracks green bean lots, roast batches, blends, recipes, and stock movements with a barrel event ledger, JWT/bcrypt auth with an owner-approval signup flow, analytics endpoints, and Excel exports.

FastAPI · PostgreSQL · React · TypeScript · Alembic · Render · Netlify · Supabase

Dashboard with follow-ups grouped as overdue, due today and this week, plus quick access to cafes and customers

1 / 4

A full-stack B2B and B2C CRM built for Vui Coffee Roasters to replace scattered notes with a fast, follow-up-driven sales workflow. The team can log wholesale cafe visits, coffee samples, customer taste profiles, and orders in seconds—while the app automatically surfaces overdue follow-ups, going-cold leads, and repeat buyers falling behind their usual order cadence.

FastAPI · PostgreSQL · React · TypeScript

Emphasis is a native macOS PDF reader built with SwiftUI and PDFKit to fix macOS Preview’s biggest pain point: annotating scanned, image-only PDFs. Instead of flattening ink onto the page, Emphasis stores free-form highlights and text boxes as real, standard PDF annotation objects—making every mark individually editable, movable, erasable, and compatible with other PDF readers. It includes a toggle between free-form and text-snapping highlights, resizable text boxes, on-device OCR search for scanned pages, Book View, and 2-up A4 export. You can also drag and drop either the annotated PDF or any single page rendered as an image straight into browsers, notes, or chat apps, with full multi-window support and native Finder integration.

Swift · PDFKit · Xcode