kukuri 0.0.1

KUKURI tries to understand any data before doing AI.

kukuri

KUKURI tries to understand any data before doing AI.

kukuri is not AutoML, and not MLOps. It is the step before all of that: the moment you have some data, a vague goal, and no time to argue about architectures. kukuri looks at whatever you have, tells you what it seems to be, and helps you get an unpretentious first baseline — in seconds, not meetings.

Don't optimize first. Verify first.

Quickstart

pip install kukuri

import kukuri

data = kukuri.load("./sensor.csv")
print(data.inspect())   # -> a DataCard: what is this data, what could it do?

or straight from the terminal:

kukuri inspect ./sensor.csv

KUKURI DataCard
========================================
Path        sensor.csv
Kind        table
Rows        1000
Cols        5
Missing     0.2%
Target?     label
Task        classification (binary)
Confidence  0.85

Notes
-----
- time-like column detected: timestamp

Philosophy

• Everything is a dataset. CSV, Excel, folders, images, waveforms — whatever comes in, kukuri reduces it to one common understanding (a DataCard) and one common API: load() → inspect() → baseline().
• A modest baseline beats a brilliant plan. When data is scarce, debating models is procrastination. Get a first answer, then decide.
• Local first. No cloud, no Docker, no GPU required. pip install is the entire setup.
• kukuri never aims for maximum accuracy. It aims for your next step.

Status

0.0.1 — pre-alpha. Today kukuri understands tables (CSV / TSV / Excel / Parquet) and takes a first look at folders. baseline() lands in v0.1. Support for more formats grows release by release — that's the game.

The name

Kukurihime (菊理媛) is the Japanese deity who mediated between Izanagi and Izanami — the god of binding things together. kukuri binds data to understanding, modalities to one interface, PoC to production, and people to AI. It also puns on the Japanese verb 括る (kukuru, "to bundle up").

License

MIT