Projects

Research and engineering work — from physics-based models to self-supervised learning.

BEV Energy Consumption Prediction

Comparative evaluation of physics-based models vs. ML (XGBoost, LSTM, TCN) for predicting energy consumption of a Sandvik LH518B battery-electric underground loader. 172k data points at 10Hz from real mine telemetry.

PyTorchXGBoostPhysics ModelingTime Series

Mining Duty Cycle Classification

Self-supervised learning for classifying mining vehicle operational states (loading, hauling, tramming, idling) from raw telemetry — eliminating the need for manual labeling across vehicle fleets.

PyTorchSelf-Supervised LearningTime SeriesContrastive Learning

Diesel Fuel Consumption Prediction

Physics vs. ML comparison for diesel fuel prediction on a Caterpillar R2900G underground loader. Collaboration with CanmetMINING (Natural Resources Canada). AI models achieved 91% error reduction over physics baselines.

PythonXGBoostLSTMPhysics Modeling

AI Scheduling App

Natural language-powered scheduling system for auto mechanic shops. Locally-run fine-tuned LLM parses scheduling requests into constraints, solved by a custom constraint satisfaction engine.

PythonRustLLM Fine-tuningOllama

augustintech.dev

This website. Built with Next.js and Tailwind CSS. Dark aesthetic, content-driven, no fluff.

Next.jsTypeScriptTailwind CSS