Work

Selected Systems

Representative engineering systems spanning multimodal learning, autonomous reasoning, neuro-symbolic AI, and embedded systems. Full implementation details live on GitHub or each program's dedicated site.

Natural Human–Machine Interfaces

Excelleve

A research program exploring thought as the next natural interface between humans and machines through non-invasive brain-computer interfaces.

Research Area
Brain-Computer Interfaces
Technologies
PyTorch · EEG · Transformers · Contrastive Learning
Outcome
79% Top-5 sentence-level imagined speech decoding across a 197-sentence vocabulary.
Deterministic Intelligence

Glazion

A neuro-symbolic reasoning system investigating how intelligent systems can continuously acquire knowledge while remaining transparent, reproducible, and trustworthy.

Research Area
Neuro-Symbolic AI
Technologies
ILP · Symbolic Reasoning · HMAC-SHA256 Audit Trails
Outcome
100% decision accuracy across 1,415 real regulatory documents, F1 score of 0.968.
Autonomous Scientific Systems

Autonomous Manufacturing Intelligence Agent

An autonomous research agent that studies unfamiliar manufacturing domains, determines how sensor data should be processed, and produces AI-ready datasets and structured knowledge representations.

Research Area
Agentic AI · Scientific Data Engineering
Technologies
Apache Parquet · JSON Knowledge Graphs · MCP
Outcome
Demonstrated on the Peregrine dataset across L-PBF, EB-PBF, and Binder Jetting processes.
Multimodal Learning

WildNet

A deep learning system for large-scale wildlife audio classification using convolutional and transformer-based architectures, developed for the BirdCLEF challenge.

Research Area
Multimodal Learning · Bioacoustics
Technologies
CNN · Channel Attention · Dual Transformer Encoders
Outcome
88% ROC-AUC across 206 bird species under ~1:500 class imbalance.
Embedded Systems

Universal Sensor Interface

Reverse engineered VEX's analog communication protocol to enable commodity sensors to operate as native VEX peripherals, reducing hardware cost by approximately 90%.

Research Area
Embedded Systems · Hardware Interoperability
Technologies
Arduino · Analog Signal Protocols
Outcome
$100 → $10 per sensor while maintaining 100% accuracy across 1,000+ cycles.