AI engineer and technical co-founder

I build AI systems and lead the teams that deliver them.

I build AI tools for industries where a wrong answer has a cost. I started on the physical side. I worked on beam metrology for radiosurgery and proton therapy, brought ML-enabled surgical tools through FDA design controls and ran quality across three manufacturing plants. Then I started writing the software. I have since built multi-agent systems that retrieve knowledge, use tools, and continuously self-improve. I have trained proprietary models for pharmaceutical production planning and deployed other models for an algorithmic trading platform. In every domain I apply one rule: develop systems to catch the things that AI must never get wrong.

B.S. Biomedical Engineering, Cal Poly San Luis Obispo. PMP.

Engineering team built and led
8 peopleAI/ML, cloud and industrial IoT
Quality organization growth
11 to 50staff across three facilities

Selected delivery

AI systems, teams and operating workflows.

Industrial maintenance

Equipment troubleshooting grounded in context

I own the technical strategy and architecture for an industrial maintenance AI startup. Its assistant combines equipment telemetry, technical documentation, hybrid retrieval and a Neo4j knowledge graph to guide troubleshooting.

I built and led an 8-person AI/ML, cloud and IoT team, taking the platform from prototype through production architecture.

Regulated quality

Medical devices under design controls

At Pro-Dex, I led new product development of battery-powered, sterilizable arthroscopic surgical tools with embedded firmware, under 21 CFR 820 and EU MDR 2017/745 design controls.

I led DHF and CE technical file remediation for legacy product families, and for a neurostimulation device startup I set up the eQMS and design controls and prepared the FDA submission.

Software adoption

Systems that reach the teams using them

At Better Nutritionals, I led NetSuite ERP, RF-Smart WMS and Qualio eQMS implementations, including training, change management and phased rollout.

I grew QA from 11 to 50 staff across three facilities and created its SOPs and training programs.

Startup and client names are omitted where the company has not agreed to be named. Team sizes and staff counts are self-reported.

Public engineering

Inspect the code and the evidence.

These projects show how I build agent applications, evaluate models and record results. The clips are screen recordings of each application running from its repository. Each results page states the revision and date of its numbers.

Retrieval and evaluation

golden-goose

Agentic RAG for medical device audit support, with revision-aware retrieval, SQL and graph routing, and a hash-chained audit log.

An eval battery gates retrieval recall, grounding, citations, sentence support, count answers, routing and reranker noise on a generated synthetic corpus. No current recorded run is published.

Chat with the agent trace, a cited answer, and the audit prep checks. No audio.
The Trace view: one record's neighbourhood in the knowledge graph, with the source document, field and confidence of each link. No audio.

Industrial IoT

obd-scanner

An agent that reads vehicle diagnostics over KWP2000 and UDS through 18 typed tools, with a pre-tool-use guard on every call.

Field test on a 2017 Infiniti Q50 3.0t, 2026-10-17: 13 modules answered, 3 stored codes, 0 broadcast frames, and the decoded EPS angle correlated at -1.00 with the ABS steering angle.

The desktop app: dashboard gauges, the steering watch with event marks, the CAN bus monitor and a 12-module trouble code scan. No audio.

Reporting

pbi-portfolio

Four browser dashboards that reproduce Power BI report features in vanilla JavaScript: slicers, cross-filtering, drillthrough pages, what-if parameters, Q&A and row-level security.

Each dashboard is one HTML page with no build step. All four share a filters pane, 10 bookmarks each and a visual header menu with focus mode, show as table and CSV export.

Three of the dashboards: a region slicer, cross-highlighting, a control chart, a what-if parameter, a Q&A visual and a View as role check. No audio.

Trading research

llm-trading-lab

Tooling for LLM trading agents, with walk-forward validation, leakage audits, a call gate and prompt change records.

On synthetic bars with two planted look-ahead leaks, the correlation screen flags one (|r| 0.6370) and the truncation test flags both.

The two review apps replaying the synthetic sample session: setup search, logged indicator panes, chart zoom, the fixed sample responses of the two placeholder models (synthetic-opus and synthetic-astra) side by side, and VWAP, EMA and MACD overlays. Prices are a seeded random walk; no model was called. Source: llm-trading-lab apps/sample-session/generate.py. No audio.

More engineering work

Experience

Engineering across software, devices and operations.

My startup and consulting roles run concurrently.

  1. Dec 2024 to present

    CTO and technical co-founder

    Industrial maintenance AI startup. Technical strategy, architecture and engineering leadership.

  2. 2024 to present

    Principal consultant and technical program lead

    Thunderbird Engineering Inc. AI development, technical program delivery and regulatory planning for medical device, manufacturing and trading clients.

    Built and deployed trading software with Python, FastAPI and LangGraph on AWS ECS, SQS and EventBridge.

  3. Jan 2022 to Jan 2024

    Senior project engineer, Pro-Dex

    Led development of Class II surgical tools, design controls and technical file remediation under FDA and EU MDR requirements.

  4. Jul 2020 to Jan 2022

    Technical project manager and quality manager, Better Nutritionals

    Quality leadership, audit preparation and ERP, WMS and eQMS implementation.

Earlier work includes biomedical engineering at Logos Systems International and construction project management at Ecostone.

Core tools: Python, FastAPI, LangGraph, Neo4j, SQL, AWS and Databricks.

Education: B.S. Biomedical Engineering, Cal Poly San Luis Obispo, 2010.

Certification: Project Management Professional (PMP), Feb 2025 to Feb 2028.

Full experience, client work and technical areas

Writing

Implementation notes and recorded limits.

Contact

Discuss a role or a project.

For senior AI engineering, AI architecture and engineering leadership roles, email me or connect on LinkedIn.

Consulting inquiries through Thunderbird Engineering are welcome.