MLOps
End-to-End Fake News Detection System
Full MLOps lifecycle with statistical ensemble (LightGBM + Random Forest + Logistic Regression), blue-green deployments, drift detection, and automated retraining.
ML / AI
Real-Time Image Classification Platform
Full-stack AI platform with no-code model training, real-time visualizations, experiment tracking, and LIME/SHAP/GradCAM interpretability tools.
🔒 Client IP — Code Private
Computer Vision
Swingaroo Golf Swing Analysis
Real-time golf swing analysis system using computer vision and pose estimation, achieving 95% accuracy in swing improvement recommendations.
🔒 Client IP — Code & Demo Private
MLOps
Serverless ML Pipeline on AWS
Architected end-to-end serverless ML pipelines using Step Functions, SageMaker, and automated model registration workflows with real-time inference at scale.
🔒 Client IP — Code Private
Computer Vision
Real-Time Multi-Camera Person Re-Identification
End-to-end system combining YOLOv8 detection, ByteTrack multi-object tracking, and ResNet-50 Re-ID to maintain identity consistency across camera views. Benchmarked on MOT17 and EPFL multi-camera datasets.
MLOps
Bitcoin Sentiment Price Prediction System
Production-grade end-to-end ML system predicting Bitcoin 1-hour price direction using sentiment analysis (VADER vs FinBERT) combined with engineered price and temporal features. Full MLOps lifecycle with automated 15-min GitHub Actions orchestration, drift detection (PSI), and rollback capability.
ML / AI
Edge AI AutoML Platform
Production AutoML system automating edge model optimization across 7 techniques (INT8/INT4 quantization, QAT, pruning, distillation) with hardware-aware simulation for 6 edge device families, cutting deployment time from 2–4 weeks to 2–3 hours.
MLOps
Iris Classification Production MLOps Pipeline
Full CI/CD/CT/CM MLOps system with automated drift detection (KS tests + PSI), hot-swappable model updates, SHAP explainability per prediction, and 115-test suite at 84% coverage. Decoupled React + FastAPI microservices deployed on Render and Vercel.
Full Stack
Luxe Hair Studio — Service Booking Platform
Production-deployed, end-to-end appointment booking system built for a premium hair salon. Features a 5-step guided booking wizard, per-stylist availability engine with race condition prevention, JWT-secured admin dashboard, and fire-and-forget transactional email, all in a TypeScript monorepo with shared Zod validation across frontend and backend.
Full Stack
Stackly — Real-Time Notification Center
Production-grade WebSocket notification system for a SaaS dashboard. Features per-user authenticated socket rooms, 7-filter notification center, per-category preferences, and a built-in live event simulator, all backed by Node.js, Prisma, and NeonDB on Google Cloud Run.
Full Stack
Knotly — Wedding Vendor Directory
Full-stack wedding vendor marketplace built with Next.js 14 App Router and Prisma. Features two authenticated portals (couple and vendor), an admin approval system, a multi-turn inquiry conversation system, Cloudinary image uploads, and a searchable vendor directory across 7 categories.
Full Stack
Pipeflow — Onboarding Flow with Feature Gating
Production-grade B2B SaaS onboarding system with a 5-step XState-powered state machine, plan-tier feature gating, and step-completion access control. Built with Vite React, Zustand, Framer Motion, and a Node/Express backend on Google Cloud Run.
Full Stack
Reforge — Conduit Rebuild & Performance Case Study
A 7-dimension performance case study: rebuilt the Conduit RealWorld app from CRA + Redux to Vite + React Query + TypeScript Strict, targeting Lighthouse 95+. Documents measurable before/after deltas across Performance, Accessibility, SEO, Core Web Vitals, bundle size, and code quality.
Full Stack
Modulus — Micro-Frontend Retail Operations Platform
A micro-frontend retail operations platform decomposed into four independently deployable React applications — Shell, Inventory, Orders, and Analytics — composed at runtime via Webpack 5 Module Federation. Each MFE ships on its own CI/CD pipeline with no coordinated releases required.
