A full-stack web application that processes audio and video files and automatically censors profane language. It combines several AI speech recognition models — Whisper, Parakeet, and Demucs for vocal separation — to detect profanity accurately even in fast-paced rap music, then applies a beep or mute using precise word-level timestamps. Users upload a file and get a clean, censored version back, or censor a live stream in real time over WebRTC. The interface shows original and filtered transcripts side by side, keeps detailed audit logs of every word detected and filtered, and provides a dashboard with processing statistics and download management. Built with React 19, TypeScript, Vite and Tailwind CSS on the front end, Django and Django REST Framework with Celery, Redis and PostgreSQL on the back end, plus AWS S3 for storage, Clerk for authentication, OpenAI Whisper for transcription and FFmpeg for media processing.
A private family office needed a live investor portal and admin console sharing one Supabase database, built entirely in Figma Make with no external IDE.
A FastAPI platform that turns property photos into cinematic videos. OpenAI Vision writes a prompt from each image, Runway generates the video.
A Python pipeline that transcribes audio recordings and separates the speakers, producing transcripts labelled by who is speaking.
An outreach system for real estate agents contacting for-sale-by-owner sellers: upload a seller list, send personalized emails, and let AI classify the replies so agents only work the interested leads.
A voice verification API: users enroll a voiceprint, then verify by speaking a random challenge phrase for liveness. Built with FastAPI and SpeechBrain ECAPA-TDNN speaker embeddings.
A cloud image gallery built as microservices on Azure Kubernetes Service, with JWT auth, container images in GHCR and CI/CD through GitHub Actions.