Applied AI · Research · Teaching
I build and study machine learning models, teach how they work, and share what I learn through research and practical projects. My interests span natural language, computer vision, and audio.
01 / SELECTED WORK
A few projects I’ve spent a lot of time thinking about. From the language we write to the world we listen to.
How do you recognize wildlife in a noisy soundscape—with only 90 minutes of CPU time? Our work brings together distillation and diverse audio models to tackle that constraint.
Listening to the Pantanal: Runtime-Constrained Distillation and Diversity-Aware Ensembling for BirdCLEF+ 2026.
The ensemble combines sound event detection models with a Perch/ProtoSSM branch using per-class rank fusion. Proceedings are currently in staging.
4th place, of 2,048 teams.
Natural language processing
PII Data Detection
A DeBERTa ensemble trained with synthetic data generated by Llama 3 70B. A data-first approach to detecting sensitive information in educational text.
Read the solution13th place, solo · of ~4,400 teams.
Natural language processing
LLM Detect AI-Generated Text
I generated 45,000 essays and trained a transformer ensemble. Trusting private cross-validation mattered more than chasing the public leaderboard.
Read the solution13th place, of ~1,400 teams.
Machine learning for education
EEDI · Mining Misconceptions in Mathematics
A Qwen retrieval and reranking cascade for identifying misconceptions. I fine-tuned the 14B retriever and built the full inference pipeline for two T4 GPUs.
Read the solutionThere’s more in the notebooks.
02 / TEACHING
I’m a Teaching Fellow for machine learning and data science at Harvard SEAS and Harvard Extension. I’ve been teaching in this field since 2016.
I like working through hard problems—and helping others find their way through them, too.
View my résumé ↗CURRENT TEACHING
AC215 · Advanced Practical Data Science
CS109A / CS109B
SUMMER WORKSHOPS · 2025 & 2026
Teaching Fellow · Summers 2025 and 2026
With Prof. Wojciech Matusik and Prof. Hanspeter Pfister
Earlier Teaching Fellow roles include Harvard Business School and these courses:
From the first model to the experiment that finally makes sense.
SHARING WHAT I LEARN
My talk on learning through Kaggle competitions.
Created 9 of the 12 section notebooks. PyTorch exercises and solutions, from neural networks to transformers.
More code to explore
My work spans applied AI, machine learning, teaching, and research, with a focus on natural language, computer vision, and audio. I enjoy tackling challenging problems and sharing what I learn.
KEEP THE CONVERSATION GOING
Machine learning, teaching, or an interesting problem.
I’d love to hear what you’re working on.
