About me
Research, software,
and people.
I’m George Sterpu, an AI engineer with a PhD from Trinity College Dublin. My background is in speech and multimodal learning. These days, I work on machine learning for building automation at Honeywell.
How I got here
During my PhD at Trinity, I researched how sound and mouth movements can help a model recognise speech. I proposed AV Align, developed Taris for online decoding, and shared the implementations as open-source projects. That research led to seven first-author papers, including follow-up journal work published in 2022.
At Xperi, I developed speech recognition for a low-power processor. At Herbst Software, I moved closer to customers, combining product management with development, DevOps, and support. At Honeywell, I now work on building models and automation.
Why forward deployed engineering?
I want to bring those experiences together: understand a customer’s problem, work through the technical constraints, build with the people who will use the system, and carry the work through to deployment.
Working in support made the human side of software much more concrete for me. I want to keep that connection as I continue working in AI.
Outside work
I’m part of Bray Runners and enjoy participating and volunteering at parkrun. Running gives me a sense of progress and community. Reading brings in perspectives from business, psychology, science, and lives very different from my own.
This site is a place for both sides: the engineering work and the things that keep me curious.
The path so far
Experience.
Jun 2025 — present
Honeywell
Senior Advanced AI Engineer
Graph neural networks for building topology, deployed to a governed endpoint on Azure Databricks. Also working on site onboarding and HVAC fault detection and diagnosis.
Aug 2024 — Jun 2025
Herbst Software
Product Manager & Support Specialist
Customer requirements, product management, software development, and DevOps for ERP-connected web software.
Feb 2021 — Apr 2024
Xperi
Senior R&D Engineer
Developed an end-to-end speech recognition engine for Ergo, running in real time with one-second latency at 100 mW. Also developed speech synthesis engines and explored model compression.
Sep 2016 — Dec 2020
Trinity College Dublin
Doctoral Researcher
Multimodal learning, audio-visual speech recognition, AV Align, Taris, and open-source research tools.