Featured Machine Learning & Generative AI Projects
AI Flyer Designer
Category: Generative AI & Design
Problem: Designing marketing flyers from scratch takes excessive manual iteration, and design software poses a steep learning curve.
Implementation: Built using an asynchronous FastAPI backend for layout synthesis paired with an interactive HTML5 Canvas engine for real-time visual adjustments.
Verified Results: Sub-second template synthesis, zero-latency client-side layer manipulation, and responsive canvas rendering.
Live Demo | GitHub Repository
Modern AI Engineer Portfolio
Category: AI Web Platform & Portfolio
Problem: Most engineering portfolios are slow static pages that fail to demonstrate actual engineering rigor or real-time model interaction.
Implementation: Engineered with React 19 and Vite featuring Lenis smooth scrolling, obsidian & champagne gold luxury styling, and a client-side AI concierge assistant.
Verified Results: 100/100 Lighthouse performance baseline, sub-second TTFB on Vercel Edge, zero runtime layout shift, and instant interactive chat responses.
Live Portfolio | GitHub Profile
Contextual RAG Assistant
Category: Generative AI & RAG Projects
Problem: Generic language models hallucinate when queried on specialized enterprise documents and lack source citation verification.
Implementation: Built end-to-end semantic retrieval with LangChain, PyTorch embeddings, high-dimensional vector databases, and cross-encoder reranking.
Verified Results: Zero-hallucination grounded retrieval on custom documentation sets, sub-250ms vector query latency, and verifiable citation source links.
Project Details | GitHub Repository
Deep Learning Auditor
Category: Machine Learning Projects
Problem: Deep neural networks silently fail during training due to train-test data contamination, vanishing gradients, or activation saturation.
Implementation: Engineered Python diagnostic modules with PyTorch and NumPy that inspect activation tensors, backprop gradient norms, and partition boundaries.
Verified Results: Automated pre-training data leakage detection, gradient anomaly alerts, and verified 30% reduction in debugging cycle times.
Project Details | GitHub Repository