Projects

I've built systems spanning computer vision, distributed platforms, embedded AI, and developer tools. These projects reflect my pattern of identifying hard problems early, designing novel solutions, and building them from concept to production.

  • Learned-prior guided text capture from video

    Built low-touch annotation pipeline producing 50k validated human-AI collaborative text annotations with limited human effort. Combined heuristics, computer vision, and vision-language models.

    Learn more

  • In-Vehicle IMU Orientation + Offset Calibration

    Novel vehicle-physics-based IMU calibration system achieving 10-25× error reduction and 50% increase in fleet coverage versus gravity-referenced method. Used vehicle dynamics as reference instead of traditional gravity-based calibration.

    Learn more

  • Online ML for Light-Based Communication

    Developed and patented computer vision and online ML algorithm enabling zero-shot optical communication from IoT devices to mobile phones, robust to outdoor lighting over distances up to 25ft. Enabled device discovery and authentication.

    Learn more

  • Edge AI DevOps Platform

    Created DevOps platform for embedded AI accelerators with version-controlled recipes and artifact management. "Docker for physical devices": device provisioning, netbooting, managing disk images from containers.

    Learn more

  • AIoT Distributed Data Science Platform

    Architected an AIoT data science platform for distributed ML training across vehicle fleet sensors. Designed an approach leveraging statistical properties for auto-discovery and reverse engineering of manufacturer-specific protocols for vehicle data.

    Learn more

  • Connected Vehicle Platform Integration Tooling

    Led software tooling development for connected vehicle platform including real-time monitoring, remote access, and anomaly detection. Created temporal cataloging and anomaly detection on CAN message periodicity and jitter.

    Learn more

  • MEMS Tablet Display Integration

    Led characterization and system integration for a novel MEMS-IGZO tablet display technology from prototype to production. Created virtual sensor via fusion of heterogeneous photodiode array as high-speed proxy of reference colorimeter.

    Learn more

  • Nuclear Safety Platform Simulation

    Led development of software simulation framework for first-of-a-kind modular safety platform. Exercised configuration, communications, control, and self-diagnostics through high-dimensional state spaces.

    Learn more

  • Nuclear Spent Fuel Pool Instrumentation System

    Drove design process and technical documentation for post-Fukushima nuclear fuel pool monitoring. Coordinated technology selection, choosing shielded time domain reflectometry. We delivered the system first to market, and most US nuclear facilities adopted our system.

    Learn more

  • Deep Learning on SuperTuxKart

    Explored CNNs, Vision Transformers, and Vision-Language Models for game AI. Implemented image classification, semantic segmentation, depth estimation, agent control, and visual question answering.

    Learn more

  • Book Corpus Text / Embedding Topic Map for Deep Learning

    Built interactive topic map from 50k book-pages using Mistral-7B embeddings, UMAP dimensionality reduction, and HDBSCAN clustering. Explored semantic structure of large text corpora.

    Learn more

  • NLP Artifacts Analysis in NeurIPS Semantic Understanding

    Investigated dataset artifacts and competency problems in a challenging NLP scenario and the limited capabilities of a small model. Fine-tuned MobileBERT, analyzed failure modes, and explored mitigation strategies for spurious correlations in training data.

    Learn more

  • Claude Code Monitor and Logger

    tui-delta wraps a TUI application (e.g. AI assistant session) to enable transparent interactive operation, while adding real-time delta processing for monitoring and logging all screen output and scrollback. Supports Claude Code, and easily configurable for other TUI applications.

    Learn more

  • uniq, but for Arbitrary Sequences

    uniqseq is like the classic Unix command uniq, but for sequences, a command-line utility for sequence tracking and deduplication. Efficiently identifies unique sequences in text or binary data pipelines and files, given extensible criteria, with flexible options for deduplication and tracking.

    Learn more

  • Pattern Database Toolkit

    patterndb-yaml is a pattern database toolkit with YAML configuration, built on syslog-ng's proven patterndb engine with patterns that are more intuitive, flexible and expressive, with added multi-line capabilities. This makes it easier to parse and normalize logs for comparison, analysis, or filtering.

    Learn more

  • LLM WebUI Launcher

    Polybolos, a customizable web UI launcher for macOS for locally-hosted utilities including LibreChat, Open WebUI, and Perplexity. Built to solve workflow friction managing multiple local AI services.

    Learn more

  • Remote Workstation Power Status and Management

    Wake-Rig is a macOS menu utility for remote wake/suspend/poweroff of Linux deep learning rigs via SwiftBar + Tailscale + Wake-on-LAN. Enables secure and efficient monitoring and power management for a remote GPU workstation.

    Learn more

  • Radio Tower Structural Health Predictive Spectral Monitoring

    Structural health monitoring system for cable-stayed radio towers using accelerometer data and DSP spectral analysis. Monitored tower cable response to wind loading, analyzed frequency spectra to track structural vibration modes and infer cable tension and component failures.

    Learn more

  • Power Analysis and Masked Logic Mitigation for Hardware Security

    Hardware cryptographic security research analyzing FPGA power consumption to identify cryptographic keys via information leakage during decryption. Implemented logic masking that reduced signal-to-noise ratio. Pattern recognition from noisy physical sensor signals.

    Learn more