We’re a small open-source group building useful AI tools - the kind you actually want in real ML pipelines. No bloated frameworks. No “soon” features. Just practical stuff, shipped.
Right now, we’ve released three libraries - each built around a real pain-point we’ve hit ourselves 👇
🐺 etsi-watchdog - “is my data drifting?”
Data drift is one of those problems that everyone knows exists but nobody monitors properly because most tools are either too heavy or too rigid.
etsi-watchdog fixes that.
- Production-first monitoring
- Static & rolling window drift detection
- PSI / KS-Test / Wasserstein built-in
- Isolation Forest support
- Slack alerts
- JSON + visual reports
- Plug-in architecture so you can register your own drift logic
Lightweight. PyData-native. Drop-in friendly.
🔍 etsi-failprint - “why is my model failing?”
Accuracy tells you how often you fail.
Failprint tells you why.
It automatically finds failure patterns across:
- Tabular data
- NLP
- Computer Vision
with:
- auto-segmentation of weak spots
- semantic clustering for text + images
- meta-feature failure analysis
- counterfactual suggestions
- clean Markdown reports
And it’s lazy-loaded, so you only pay for the dependencies you actually use.
🌋 etsi.etna - a tiny neural-net library for structured data
Most ML tools force a choice:
simple or fast.
Etna says: why not both?
- Rust-powered core
- Python-first API
- auto task detection (classification / regression)
- auto preprocessing
- built-in metrics
- MLflow tracking by default
Perfect for quick experiments, teaching, and small-scale production.
🧠 Why we’re building this
We just want AI tooling that feels:
- simple
- transparent
- fast
- well-designed
- actually usable in production
So we’re making it — in the open.
⭐ If this sounds like your vibe
Check out the libs, try them, break them, open issues, contribute, hang out.
We’re building. And we’re having fun doing it.
More soon. 🚀
— etsi.ai
Top comments (0)