SimonPark
Scroll
01Impact
Measured in production.
- −60%Snowflake costBlock
- +30%Detection accuracyNexxa AI
- −50%Prediction errorStern Laboratory
- +32%Test coverageDandelion Networks
01 / 04
02Publication
Nuclear Engineering and Design
An Artificial Neural Network (ANN) Model to Predict Critical Power in a CANDU Fuel Element Simulation
Read the paper03Experience
Where the models shipped.
S/P
- Goose agents on Snowflake + Looker MCPs, −70% investigation time
- Snowflake compute −60% (~$20K/yr)
- Async Prefect flows, +50% faster
- Fine-tuned YOLO-NAS / Detectron2 / EfficientDet, +30% accuracy
- Docker + AWS deployment
- llama.cpp tuning, +20% inference speed
- NumPy-only CHF neural network, −50% error
- PySide6 + SQLite3 viewer shipped with PyInstaller
- Moved Validation Pool data to BadgerDB with prefix hashes
- +32% test coverage (Ginkgo, Gomega)
- Fixed 50+ microservice issues (SQL, Docker, Postman)
- Migrated the ReadyAPI licence server
Hover a role for details
© 2026 Simon ParkLand data: Natural EarthBuilt from particles