Building AI systems, data products & full-stack apps.

AI-powered legal assistant web application that provides intelligent legal query responses, document analysis, case summarization, contract compliance checking, and advocate recommendations using retrieval-augmented generation (RAG) with vector search and generative AI.

A production-grade full-stack rental application built for a job assessment, featuring an ultra-minimal editorial luxury aesthetic. Built with Next.js, TypeScript, and Prisma with feature-based modular architecture. Includes authentication, messaging page, cursor-based pagination, and a sophisticated design system with Tailwind CSS v4, shadcn/ui, and Framer Motion animations.

MLOps confidence collapse monitoring system, a production-style ML pipeline built around a FastAPI inference service, with MLflow experiment tracking, Dockerized deployment, Jenkins/GitHub Actions CI/CD, and live Prometheus + Grafana monitoring. The project investigates "confidence collapse": as input noise increases, the model's accuracy drops from 97% to 53%, a failure mode the monitoring stack is designed to surface.

Fomi AI Studio is a premium AI content-generation interface I built for a frontend technical assessment, where the task was to recreate a provided design mockup as a fully functional, responsive Next.js application. I delivered a pixel-faithful recreation of the required design and pushed further with an original second "concept" design, both built on Next.js (App Router), React, and Tailwind CSS v4. Generation is simulated through a mock API so the complete experience — prompt composer, style presets, model and settings controls, image and video output, a gallery with lightbox, history, and full loading/empty/error states — works end to end. Under the hood it runs on a token-driven design system that powers a light theme, a dark mode, and a distinct cinematic theme from one shared set of CSS variables, with custom accessible components, responsive layouts down to 320px, and motion that respects reduced-motion preferences.