Research software and systems I’ve built or contributed to, mostly around real-time multimodal data collection for learning analytics.
2025 – present
Creator and lead engineer
A math assessment and AI-tutoring platform embedded in the OECD's PILA and deployed in a randomized controlled trial across hundreds of schools in France, Germany, the Slovak Republic and Latvia. It includes an LLM tutor with learner-model context, LLM translation and knowledge-graph pipelines, xAPI logging and a real-time teacher dashboard.
SvelteKit, TypeScript, LLMs, xAPI, RCT
2023 – present
Creator and lead engineer
An open-source WebRTC platform for real-time multimodal classroom data. It sustains 200 concurrent streams at about 200 ms latency and has archived 400+ hours (more than 10 TB) for 1,000+ researchers and students. It is deployed in classroom studies in Indiana, Tennessee and North Carolina.
Rust, Next.js, WebRTC, LiveKit, Tauri, AWS ECS
2022 – 2023
Co-developer
A distributed streaming framework for real-time multimodal data retrieval and processing, used by 100+ people in Nashville-area classroom studies. I built its orchestration layer, SvelteKit dashboard and pipeline library (video, audio, people tracking, live transcription and video anonymization).
Python, ZeroMQ, SvelteKit, WebSockets
2025 – present
Lead engineer
Ongoing work on live feedback for classrooms: speech recognition with NVIDIA Riva and AWS Transcribe, computer vision models on edge and cloud (NVIDIA Jetson, AWS, Jetstream2), and on-device LLM summarization of live classroom transcripts.
NVIDIA Jetson, Riva, Computer vision, On-device LLMs
2019 – 2022
Core contributor
An NSF-funded visual environment for building and running deep learning pipelines (700+ GitHub stars). I led its Keras parser modernization and Svelte web components, and used it to reproduce a CNN-based photometric redshift study on SDSS galaxy images.
JavaScript, Python, Keras, Svelte
2019 – 2022
Lead developer
GMSO is the core topology library of the NSF-funded Molecular Simulation Design Framework (MoSDeF). I led its development, created forcefield-utilities, and set up CI, Docker publishing and release tooling across MoSDeF libraries.
Python, Molecular simulation, CI/CD
2019 – 2022
Developer
Tools for DARPA's Symbiotic Design for CPS program: Circuit Maker, a web-based circuit design studio that uses deep learning to predict the next likely component; a Python client for scripted UAV design evaluation; and Fusepack, a genetic-algorithm component placement tool for UAV fuselages.
Deep learning, Web, Optimization, Docker
2022 – 2024
Research engineer
Deep active learning and surrogate models for underwater vehicle hull design, driven by a combined CAD–CFD simulation pipeline.
Deep active learning, Surrogate modeling, CFD