CRONOS: Time-series analytics without training — what would you use it for?

Hey everyone!

We're working on CRONOS — a time-series analytics engine that works without training data.

The idea: instead of learning patterns from historical examples, we compute mathematical invariants directly from signals. Same engine works across completely different domains.

What it does:

• Anomaly detection

• Trend prediction

• Multi-stream correlation

• Event detection

Key specs:

• 122 nanoseconds per analysis (500,000× faster than LLM)

• Runs on ARM Cortex-M4, no GPU

• 100% deterministic — same input, same output, always

• Zero-shot — works from the first sample

Validated on public benchmarks:

• Industrial: 95.9% F1 on bearing faults (CWRU)

• Medical: 100% EEG seizure detection (CHB-MIT)

• Energy: 99.9% load disaggregation (REDD)

• Space: 99.3% exoplanet transit detection (Kepler)

Trade-off: we typically see 3-5% lower accuracy than supervised models trained on domain-specific data. But you can deploy immediately without collecting failure examples.

We'd love to hear:

• What time-series problems are you solving today?

• Where does "not enough training data" block your projects?

• What would you need to see to trust a zero-shot approach?

Product page: https://nextx.ch/cronos

Live demo (same math engine): https://engine.aqea.ai/ui

AMA about the tech, benchmarks, or limitations!