Behavior ML Pipeline

Overview

I lead a multi-year project at HHMI Janelia studying goal-directed navigation in adult Danionella, tiny, nearly transparent fish. No hardware, training paradigm, or analysis pipeline existed for this work, so we are building all of it from scratch.

My contributions span hardware, software, and data science:

• Co-designed custom hardware with engineers: automated feeders that dispense 1-5 brine shrimp eggs per reward (pressure-regulated reservoir, vision-based egg detector, syringe-pump dispenser), multi-camera mounts, and IR illumination
• Designed behavioral experiments, including audio and projector-based visual cues that guide fish toward target zones
• Contributed to the MATLAB experiment control software and Arduino firmware that synchronize cameras, audio cues, projector stimuli, and feeders in real time
• Built semi-automated Python pipelines that ingest and process 100+ GB/day of multi-camera video, integrating deep learning models for detection and tracking
• Refined tracking and detection output and applied statistical analysis to characterize learning trajectories across animals and sessions
• Co-authored a public Quarto documentation site so others can reproduce the system

I presented this work at the 2025 Danionella Workshop. It was also presented at Society for Neuroscience 2025, in combination with related team work. The poster below lists the full collaborator team.

Open the poster in full screen

Skills

Python · MATLAB · Deep learning detection and tracking · Data pipelines · Scientific communication · Project leadership