All case studies
ResearchApplied ML · Forecasting2025 — present
Chronos
Time-series forecasting toolkit that ships calibrated uncertainty by default, not point estimates.
Role · Lead engineer
Overview
A forecasting library that ships well-calibrated intervals — not just point estimates — by default.
Problem
Most forecasting code emits point predictions; production systems need honest uncertainty to make decisions against.
Solution
A small set of well-tested probabilistic models with consistent APIs and calibration diagnostics built in.
Tech stack
- Python
- PyTorch
- Probabilistic models
Engineering notesWhat's shaping this build.
Design goals, philosophy, planned architecture, and where the project stands today. No fabricated benchmarks — only what's actually driving decisions.
Engineering notes
What's shaping this build.
Design goals, philosophy, planned architecture, and where the project stands today. No fabricated benchmarks — only what's actually driving decisions.
Design goals
- Calibrated intervals as a default output, not an optional flag.
- Consistent APIs across model families.
- Calibration diagnostics as first-class model outputs.
Philosophy
A point forecast without an interval is a decision waiting to go wrong.
Planned architecture
PyTorch-based probabilistic models behind a shared fit/predict/calibrate interface, with diagnostic reports produced on every fit.
Current stage
Model surveys and API design.
Links
Repository and demo links will appear here once the project is ready for public review.