Systems I have helped build, operate, or translate into working research environments: cloud neuroscience workflows, behavioral experiment infrastructure, electrophysiology analysis systems, and real-time creative hardware.
The throughline is practical translation: taking messy signals, lab constraints, hardware, and human workflows, then shaping them into systems other people can actually use.
system record
brief available
RetinoAffective: Falsification-First Brain Encoding of Naturalistic Video
A falsification-first research harness for predicting fMRI responses to naturalistic video. Inspired by Meta's TRIBE, but scored on region-specific, noise-ceiling-normalized, control-matched prediction behind preregistered gates — producing a cross-dataset-confirmed visual encoder, a set of well-powered negative results, and a characterized failure mode.
neurOS-v1: A Modular Operating System for BCIs and Neural Foundation Models
A 10-package Python monorepo that carries neural data end-to-end — hot-swappable device drivers across 16+ biosignal modalities, agent-based real-time orchestration, a registry-driven model zoo, the neuroFMx multimodal foundation model, mechanistic-interpretability tooling, and cloud-native streaming, serving, and observability.
Role
Author and architect
Scope
Real-time neural data streaming, processing, classification, and multimodal foundation-model R&D
neurOS-v1 Package: neuros-neurofm — the neuroFMx Foundation Model
The from-scratch multimodal foundation model of neurOS-v1: per-modality tokenizers, a Mamba state-space backbone, self-supervised masked-modeling objectives, and LoRA + unit-ID adapters for cross-session transfer.
The runtime heart of neurOS-v1: an asynchronous agent orchestrator, the pluggable processing pipeline, CV plugins, and adaptation/health monitoring that turn streamed signals into real-time inference.
neurOS-v1 Package: neuros-foundation — Reference Foundation Models
Published neural foundation models — CEBRA, NDT, POYO, Neuroformer — behind one base interface, with dataset loaders, so external architectures are directly comparable inside neurOS-v1.
A unified, hot-swappable driver API across 16+ biosignal modalities, plus NWB/Zarr I/O, so any device plugs into neurOS-v1 behind one contract without changing the pipeline.
Role
Author and architect
Scope
neurOS-v1 monorepo package
Stack
neuros-v1 / drivers / adapters / nwb / zarr
Context
BCI & Real-Time Systems / Neural Data Infrastructure
neurOS-v1 Package: neuros-models — Model Zoo & Registry
A registry-driven library of deep and classical models behind one interface, so architectures are swappable and directly comparable, with a SageMaker launcher for cloud training.
neurOS-v1 Package: neuros-cloud — Streaming, Storage & Federated Training
The distributed backbone of neurOS-v1: message-bus streaming (Kafka, Redis, ZeroMQ), lakehouse storage and export (Iceberg, Petastorm, WebDataset), federated learning, and LSL sync for real deployments.
The human surface of neurOS-v1: a FastAPI serving API, a Streamlit dashboard, and a CLI to run pipelines, benchmarks, and the Constellation demo without writing code.
Role
Author and architect
Scope
neurOS-v1 monorepo package
Stack
neuros-v1 / fastapi / streamlit / cli / serving
Context
BCI & Real-Time Systems / Neural Data Infrastructure
A focused service that weights heterogeneous data sources for training mixtures, so a foundation model learns from the right blend of data rather than from raw volume.
The reward-certainty paradigm at NEATLabs: reward contingencies flip unpredictably while rats must relearn which action pays off. Beta and high-gamma oscillations track reward valence and probability, and optogenetic beta-frequency stimulation causally perturbs adaptive behavior. Anchors the Journal of Neuroscience paper on reward certainty.
Role
Lead Lab Programmer — behavior systems, LFP analysis, RL-model support
NEATLabs Paradigm: Delay Discounting — Reward Value & Subjective Value
The reward-value paradigm at NEATLabs: rats choose between smaller-sooner and larger-later rewards while reward-locked cortico-striatal beta oscillations track reward magnitude, delay cost, and computationally modeled subjective value. Anchors the Cognitive, Affective, & Behavioral Neuroscience paper on beta as a reward signal.
Role
Lead Lab Programmer — behavior systems, time-frequency & value-model analysis
Supporting a public Sabatini Lab DataJoint workflow for organizing and processing multimodal neuroscience sessions across imaging, photometry, behavior, electrophysiology, and DLC video data.
Role
DataJoint deployment and operations support
Scope
Public workflow repo, lab data organization, DataJoint setup, notebook-driven analysis, SciViz testing
Operational support for DataJoint-based neuroscience workflow infrastructure connected to Allen Institute Mindscope-style large-scale datasets, with emphasis on reproducible processing, schema organization, cloud execution, and reviewable outputs.
Role
DataJoint workflow operations and deployment contributor
Scope
Large-scale neuroscience workflow support, DataJoint Elements integration, cloud-oriented processing patterns
Research systems work at UC San Diego spanning rodent behavioral paradigms, Raspberry Pi operant boxes, large-scale LFP analysis, TBI studies, drug-condition comparisons, and publication-supporting neural dynamics pipelines.
Role
Lead Lab Programmer and neuroscience research contributor
Scope
2017-2022, behavioral task systems, LFP/behavior analysis, Raspberry Pi operant boxes, publication support
Stack
neatlabs / raspberry-pi / matlab / lfp / behavior
Context
NEATLABs Research / Experimental Systems / Neural Signal Discovery
The behavioral-inhibition paradigm at NEATLabs: rats act, withhold, or wait in response to cues while distributed cortico-striatal LFP and human EEG reveal separable action and inhibition networks. Anchors the eNeuro large-scale-network paper and the rodent default-mode suppression study.
Role
Lead Lab Programmer — task systems, event extraction, LFP/behavior analysis
Scope
Go/No-Go and Go/Wait behavioral inhibition, multisite LFP, human EEG comparison
Creative hardware systems connecting NeuroSky-style brain input, Raspberry Pi control, audio-reactive LED strips, NanoLeaf-style lighting, and lagless audio routing for synchronized real-time visual feedback.
Role
Designer and implementer for class and personal creative hardware systems
Scope
NeuroSky to Raspberry Pi LEDs, dancyPi-inspired audio LEDs, NanoLeaf-style lighting, galaxy light, split audio routing
Stack
raspberry-pi / neurosky / led / audio-reactive / real-time
Context
Real-Time Creative Systems / BCI and Signal Feedback / Hardware Prototyping
Behavioral inhibition and impulsivity paradigms where cues, waits, premature responses, omissions, rewards, and outcomes had to become reliable neural-alignment events.
Changing-reward tasks used to study certainty, reversal behavior, high-gamma activity, and cortico-striatal network updates.
reward certaintyreversalhigh gamma
Raspberry Pi Behavioral Boxes
Operant-box infrastructure and MATLAB/Simulink workflows for controllable rodent behavior, synchronized events, and electrophysiology-ready experiments.
raspberry pioperant boxessync
Experimental comparison space
Across these systems, the analysis work had to support comparisons across healthy and TBI-related animal cohorts, task conditions, trial outcomes, and pharmacological contexts including saline, methylphenidate, and ketamine where relevant to the lab's experiments.