AI that works
in the real world.

Selected work is patent pending 🔒

Built an automated RAG intake pipeline using Microsoft Azure, service accounts, and multiple machine learning models. The pipeline inspects documentation across cross-functional engineering teams for discrepancies and automatically recompiles the knowledge base after verification.

Developed custom tooling and agent skills connecting technical retrieval with engineering workflows for power, latency, thermal, and routing calculations.

Created a custom analytics tracker for an internal React-based chip design platform using MSAL enterprise authentication. Instrumented page usage and application interactions to understand how engineers use the platform.

Extended the tracker with LLM and agent analytics, including agent usage, time to first token (TTFT), semantic similarity between questions, and workflow telemetry. Built SQL-backed analytics and a usage dashboard to guide improvements to the platform and its engineering assistant.

Led senior and principal engineers in Arm’s San Diego office through a hands-on ESP32 and e-ink build workshop, guiding participants as they turned connected displays into custom desk tools.

Participants could build a stock ticker, connect Microsoft Teams to display their next meeting, show what was playing on Spotify, or take the hardware in their own direction with a custom build.

Built dynamic documentation for an internal Arm platform that adapts to the page a user visits the documentation site from, bringing relevant guidance into the context of their current task.

Organized agent tool documentation separately from platform actions and quick tips, helping users find both detailed tool guidance and practical instructions for navigating the platform.

Patent pending / in progress

Developed machine learning models using real sensor data from Airbuild’s microalgae wastewater treatment bioreactors in Green River, Utah. Paired biomass growth prediction with algae-health anomaly detection to forecast growth and identify signs of declining health.

Patent pending / in progress

Connected Airbuild’s AWS data stores to a website I designed and built for a real-time view of algae health. Brought sensor readings into a live dashboard so the team could monitor conditions in one place.

Connected the same data to our agents, enabling the team to ask about algae health and get answers grounded in live readings.

Coming soon: Automated nutrient dispensing to adjust algae nutrient levels when health declines.

Architected and deployed an internal multi-agent workforce for team members. During onboarding, the system captures repetitive operational tasks and dynamically equips agent instances with approved enterprise skills, calendar/scheduling connectors, Canva automation, and live IoT sensor telemetry.

Designed, architected, and shipped a full-stack project management platform from zero — currently in active company-wide beta testing. Built reactive user interfaces, implemented real-time database sync with Firebase, and integrated an internal LLM trained on architectural drawings and technical documentation.

Patent pending / in progress

Developed Airbuild’s second-generation sensor panel, contributing to component selection, hardware assembly, firmware code, and software. Enabled over-the-air (OTA) updates so the panel could receive updates remotely.

Built Airbuild’s company website and have complete ownership of its ongoing development and maintenance. Visit Airbuild ↗

Engineered and deployed an executive market intelligence platform used across North American Marketing and Insights teams (brands including Godiva, Flipz, McVitie's). Created an automated ingestion service allowing non-technical stakeholders to email reports to a service account for automated dashboard parsing, paired with AI-driven competitor release synthesis that directly shaped product roadmaps.

Coordinated 10+ Python research projects while advancing deep learning code retrieval models under Prof. Vivian Huangfu. Enhanced neural code search representations by combining domain-specific LLM pre-training objectives with semantic code embeddings to improve developer retrieval accuracy across large codebases.

Standalone WiFi-enabled wearable fall detector built with an Arduino Nano ESP32 + MPU6050 6-axis IMU. Uses an orientation-independent Signal Magnitude Vector (SMV) threshold algorithm running locally on-device — achieving sub-2-second alert latency via Discord webhooks. Built for under with no recurring subscription fees. Read Research Paper ↗

Researched custom GPT learning assistants grounded in master’s-level course syllabi and assignments, documenting the work in a research paper. Read Research Paper ↗

As the sole undergraduate Teaching Assistant for graduate systems courses (MIS 695 & 515), developed Python autograding scripts that cut evaluation time by over 70% while improving grading consistency across 30+ graduate students.

Directed SDSU's flagship AI hackathon (sponsored by Anthropic & Zip Launchpad) and founded the CodeBlitz competitive programming series. Authored novel algorithmic problem sets, built automated Python test runners for instant leaderboard scoring, and led logistics for hundreds of participating students.

Built a custom spin class planning platform for a spin studio, giving its indoor cycling instructors tools to plan and deliver complete rides. Plan complete ride profiles with RPM ranges, intensity levels, and move durations with a real-time total duration tracker. Integrates Spotify and SoundCloud Web APIs for track search and real-time audio BPM detection, with one-click export. Launch Live App ↗

Built an agent for a real estate client to track clients, the homes they’re following, and when the realtor last reached out. The system keeps client relationships and property interests organized in one place.

Added proactive follow-up so the agent reaches out to clients when the realtor misses a check-in, helping maintain contact without relying on manual reminders.