We build systems that are provably
correct, reliable, and fast.

Problem  ·  Operational data from multiple Cenovus facilities (sensor readouts, raw logs) was arriving with irregular sampling, missing values, and inconsistent units, blocking any reliable forecasting or predictive-maintenance work.

Outcome  ·  Built a secure end-to-end pipeline covering ingestion, feature engineering, model training and reporting, with gradient-boosted trees and recurrent neural networks evaluated per task. Models were containerised and deployed to cloud; daily dashboards now surface insights for engineers and automated reports replace manual rollups.

Problem  ·  Detecting Alzheimer's disease from hyperspectral imaging on blood samples. A noisy time-series problem with real risks of data leakage and outlier-driven results that would invalidate the model.

Outcome  ·  Built the pipeline with Gaussian smoothing, MinMax scaling and MiniRocket-style feature extraction, then benchmarked GRU, RNN, LSTM and HIVE-COTE 2.0 against XGBoost, LightGBM and Random Forest. Random Forest came out on top.

Problem  ·  Teachers want interactive lessons but have almost no time to build them. Turning a topic line like "make a counting game for kindergarten" into a polished, classroom-ready activity is out of reach in a normal prep window.

Outcome  ·  Shipped a generation pipeline: Claude 3 Sonnet decomposes the request and drafts the activity, a Rust Actix-Web service moderates, spell-checks and sanitises the output, and a Next.js frontend streams the activity live to the browser for preview and edits. Embeds directly into SMART Boards and LMSs.