Sam Harrison

Software Engineer — Databricks (Previously Neon)

Nov 2024 - Present

San Francisco, CA (Remote)

Machine Learning Engineer II — Skyworks Solutions Inc.

Jul 2024 - Nov 2024

Ottawa, ON

• Developed physics-informed neural networks of GaAs pHEMT devices, improved bias point selection accuracy leading to reduced intermodulation distortion and simulation wall times
• Implemented a PyTorch-to-Verilog-A transpiler that converts trained neural networks into Verilog-A modules, enabling device teams to drop deep learning models straight into Cadence and Keysight simulators
• Created a CLI for HPC management that simplified ML workflows through templated resource allocation, automated storage mounting, containerized Apptainer environments, and experiment monitoring

Algorithm Developer Intern — GBatteries

Aug 2023 - Jan 2024

Ottawa, ON

• Developed CNN-LSTM and LightGBM models for Li-ion battery state estimation, improved accuracy by 4% and enabled inference on resource-constrained edge devices
• Created a real-time inference system using FastAPI and Redis telemetry data buffering, delivered low-latency state predictions during active charging cycles for hundreds of batteries
• Consolidated battery data from multiple charging platforms into a central MongoDB database, and developed a React web platform for analysis of battery cycling and EIS results

Junior Data Scientist — Public Services and Procurement Canada

Apr 2022 - Sep 2022

Quebec, QC

• Developed an automated review system for the National Project Management System to provide early warnings for projects at risk of exceeding time, budget, or scope constraints, reducing the quarterly review timeline from over 60 hours to approximately 5 minutes
• Created a comprehensive dashboard to visualize project data, highlighting trends across projects, regions, and project managers

FLood2 — Undergraduate Research in CFD | blog.samharrison.ca/posts/flood2

• Novel, highly parallel turbulence characterization algorithm designed for execution on HPC clusters, and applied to DNS results from Nektar++ simulations. Implemented in C++11, Python, and Rust

pgtensor | github.com/sam-harri/pgtensor

• Postgres extension written in Rust adding a tensor data type and an ONNX inference engine for in-database model execution using background worker processes and shared memory; supports multiple dtypes and shape validation

TensorCraft | tensorcraft.click

• Drag'n'drop neural network builder with real-time feedback on tensor shapes and compilation to PyTorch implementation

Engineering Capstone — Hatch Ltd.

• Collaboration with Hatch Ltd. to design a uranium milling process for a pilot plant in Saskatchewan, 1st place at University of Ottawa plant design competition

Languages: Python, Rust, Typescript · ML/DL: PyTorch, SciKit Learn, Polars, Pandas, NumPy, ONNX · Databases: SQL, Postgres, MongoDB, Redis · DevOps: Docker, Compose, Git, CI, Linux, AWS · WebDev: React, Next, Tailwind, Drizzle