• Created notification pipeline which gathers events published to MQTT topic in Amazon SQS, and uses a Golang consumer with the Apprise CLI in order to send out over 10,000 notifications per second.
• Designed and implemented a full-stack no-code pipeline builder using HTML/CSS/JavaScript and Python Flask. The application converts user-defined visual workflows into YAML and processes them through a Flask /run API route to execute workflows locally.
• Supported development of an end-to-end media streaming pipeline that ingests any OpenCV-compliant video source, converts it to RTSP via FFmpeg, and processes it through a media server to output WebRTC or HLS. Developed a frontend viewer using HLS.js and native WebRTC support to stream content in real time.
• Leveraged above pipeline to develop a scalable full-stack application capable of streaming up to 20 simultaneous OpenCV-compliant video inputs with real-time inference overlays. Implemented support for running different models per input and achieved an overall throughput of 600 FPS.
• Contributed to application which uses real time AI inference results from user’s webcam and maps facial embeddings to 3d model, invoking a “v-tuber” like experience.
• Developed plugin for open source project Frigate, removing over 80% of setup time for using models with Frigate, and allowing access to the over 1000 models available on DeGirum’s AI Hub.
• Created tests for Degirum’s Javascript AI SDK using Jest, including fully mocked websockets, javascript canvas, and model inferences.
• Created multiple CI/CD actions on Github to generate documentation, perform releases, and run tests.