I build open-source software for industrial water treatment and for search evaluation. I try to publish things a stranger can understand, run, and check.
cooling-tower-chem — tells you whether water will form scale or corrode pipes. The classical water-stability indices as a dependency-free Python library, every formula cited to the paper it came from. The nearest open-source neighbour is tidywater on CRAN (R, drinking water); no installable Python package ships this index family. On PyPI, 112 tests, MIT.
TGF — advisory water treatment for industrial cooling towers: risk indices, anomaly detection, forecasting, and dose recommendations. In advisory use at eight cooling towers across two Indian plants. A human authorizes every dose. Apache-2.0.
Neural E-Commerce Search
— necs-validate, which checks TREC evaluation files for structural errors before
anyone reports a score from them, as a CLI, a GitHub Action, and a browser page.
Plus a conformance corpus for diffing trec_eval implementations, and a
retrieve-and-rank reference for Amazon ESCI. MIT.
- Retail Sales Forecasting Engine — forecasting benchmark on the M5 dataset.
- The Recommender — state-space sequence models compared against SASRec-style recommendation.
- Social Media Event Organizer — a Java Swing group project, with 71 JUnit tests and team attribution.
- SAiDL Spring Assignment 2025 — coursework on state-space models and CoreML.
Reproducible experiments — configs, seeds, raw results, stated evaluation protocols. Honest boundaries between a prototype, a simulation, and a deployed system; where I have overstated something, I have withdrawn it and said so.
If you reproduce a result or find a failure case, please open an issue in the relevant repository.