- Internal
Hybrid AI Product Innovation Portfolio
May 2025 - May 2026
Associated with AI Product Engineer & Innovation Lead (Dell Technologies)
Built an AI Product Innovation (R&D) portfolio from zero, with over 35 feasibility prototypes and a Kanban team of 10. It led to 22 patent applications with me as first-named inventor, and many of its concepts ship today as production components.
Key Learnings
- How to filter ideas against a company's strategy, and when to kill a concept early.
- How to turn a rough idea into a feasibility prototype fast enough to decide whether it's worth pursuing.
- How to write business cases and PRDs that earn leadership funding.
- How to plan research in phases, so each stage earns the next one.
- How to run a self-managing group across time zones.
- How research actually lands in product roadmaps, and why the handover matters as much as the idea.
- How the patent process works, from invention disclosure through internal review to filing.
- Together, these let me take an idea from first sketch to a funded, prioritized backlog item, and know when to stop.
- Commercial
Hybrid AI Platform
May 2024 - Apr 2025
Associated with AI Software Developer & Delivery Lead (Dell Technologies)
As sole developer, designed, built, and owned 7 OpenAI-compatible AI services (speech recognition, SLM chat, and function calling) on ONNX Runtime, running on the NPUs and GPUs of heterogeneous AI PC fleets. They shipped to every Dell AI PC customer on two hardware platforms and were featured at CES and DTW.
Key Learnings
- How language models run inference: model parameters, sampling strategies, and decoding loops.
- How prompting techniques shape a model's output, and why hallucinations happen.
- How prompt injection works, and why untrusted input has to be kept apart from instructions.
- How context windows and tokenization limit what a model can take in, and what it costs to run.
- Why more context isn't always better: longer inputs are slower and costlier, and can bury what matters.
- How quantization and model size trade off against accuracy.
- How latency differs between NPUs and GPUs across heterogeneous device fleets.
- How to tell whether a model is good enough to ship, using benchmarks such as MMLU and WER.
- Together, these let me scope AI features realistically and weigh speed, accuracy, and cost trade-offs before anything is built.
- Internal
Web Platform Modernization
Aug 2021 - Apr 2024
Associated with Junior Software Developer, then Software Developer & Delivery Lead (Dell Technologies)
Built from scratch, together with the team, the .NET Web API services behind 3 operator-facing modules that replaced legacy components. From release, they were used by all of Dell's factory engineers across the Americas, EMEA, and APAC.
Key Learnings
- Why replacing legacy systems piece by piece keeps disruption low for the people who use them.
- How API decisions ripple out when 6 other teams depend on your services.
- Why performance, reliability, and monitoring matter as much as features once software is in production.
- How CI/CD and automated testing turn releases from risky events into routine ones.
- What helps interns and juniors get productive quickly.
- Together, these help me spot delivery risks and dependency problems early.