• Client had no single tool to manage their digital marketing, relying on disconnected platforms. Built a production SaaS platform from scratch with a React + TypeScript frontend, Node.js/Express backend, and PostgreSQL, handling end-to-end billing via Stripe and AI features via OpenAI.

• Owners had no visibility into why their restaurant was underperforming online across search, reviews, and social.

Engineered an AI-powered audit engine that analyzes Google Business, local SEO, review sentiment, social media, and competitors, surfacing a prioritized action plan in a single dashboard.

• Menu optimization required expensive consultants and days of manual analysis. Built an OCR-based menu ingestion pipeline with OpenAI extraction generates recommendations, and feeds a Canva-integrated design studio, cutting analysis time from hours to minutes.

• Designed and shipped interactive graph-based dashboards for FTU anatomical data exploration using React and

AngularJS backed by Node.js on AWS, reducing researcher navigation time by 3x over the prior interface.

• Tightened RESTful API integrations with data normalization and validation, fixing recurring client-server mismatches and bringing frontend error rate down to under 0.1% across 13+ months of production use.

• Automated segmentation pipeline – Built an OpenCV + NumPy system to segment biomedical images, reducing manual effort from hours to minutes. Scaled processing to 400k+ images using CUDA + multithreading, cutting runtime by 40% while preserving image quality via adaptive resizing and aspect-ratio alignment.

• Anatomical visualization system – Developed a Python-based engine to auto-generate SVG schematics of organ blood flow from structured CSV datasets. Implemented ID normalization, FTU to FTU vascular inference, and custom Matplotlib rendering utilities with adaptive layouts, producing high-resolution, reproducible biomedical diagrams.

• Built the full-stack sentiment analysis product: React dashboard consuming Flask and FastAPI endpoints backed by

Dockerised PyTorch models, raising transformer accuracy by 15% and throughput by 10% across 5+ units.

• Designed a RESTful API layer with Docker and Flask to decouple ML inference from the frontend, allowing each service

to be updated and scaled independently without breaking the UI.

• Launched a Kubernetes-based recommendation system with Prometheus and Grafana monitoring, achieving a 20% boost

in personalization efficiency and 25% higher anomaly detection accuracy.

• Built a full-stack object detection platform: React frontend and Flask REST API serving a DETR model with a ResNet backbone, achieving 90% detection accuracy and meeting client requirements on the first production deployment.

• Configured KubeEdge clusters on WSL with Kubernetes to deploy lightweight AI models to edge devices, sustaining 81% model accuracy at minimal latency in low-connectivity environments and built a React dashboard for live inference monitoring.

• Optimized Docker image builds and deployment scripts, reducing inference response time by 15% across all target platforms.