Namtheg AutoML

نَمذِجNamtheg

From a raw table to a live model

Namtheg studies your data like an analyst, trains every candidate model on an NVIDIA H200, explains the winner with numbers it actually computed, and serves it for live predictions.

How it works

  1. 1Upload a table

    CSV, TSV, Excel, Parquet or JSON, up to 2 GB. Pick the column to predict.

  2. 2The analyst investigates it

    An AI analyst runs code on your data in a sandbox: what each column means, leakage, gaps and the cleaning it needs.

  3. 3Models train on an NVIDIA H200

    Boosted trees, SVMs, nearest neighbours and linear models, cross-validated in parallel, then the best one is tuned.

  4. 4Results with a grounded explanation

    The champion, its scores and the full leaderboard, explained only with numbers the run computed.

  5. 5Deploy for live predictions

    One click puts the model behind an HTTPS endpoint, with a test page and a QR code to share.

Upload

CSV, TSV, Excel, Parquet or JSON

Drop a dataset here

Checked on upload, then profiled on the server.

Download a sample (insurance.csv)

Configure run

Pick the column to predict. The rest is detected automatically.

Waiting for a dataset
Upload a file to choose a target and start training.

Data profile

Completeness and column types

No profile yet
Completeness and column types appear after upload.

Dataset preview

Raw rows and per-column statistics

Nothing to preview
The first rows of your dataset show here once it is uploaded.

Pipeline

Profile, detect, explore, engineer, train, report

No run yet
Configure a run on the Data stage and start training.

Run diagnostics

Recorded run state. Training runs on separate GPU containers.

Nothing recorded
Stage, hardware and timing show here while a run trains.

Notebook

One cell per pipeline stage

The notebook is empty
Each pipeline stage runs as a cell here, with its code and log.

Champion model

The model the pipeline picked, and why

No champion yet
Appears when training finishes.

Score

The selection metric on held-out data

No score yet
Appears when training finishes.

Report and files

The report, every graph, the data and the model

Nothing to download yet
The report, cleaned data and model package appear here.

Graphs

How the champion does on held-out data

No plot yet
Appears when training finishes.

Feature importance

Which columns drive the prediction

No features yet
Appears when training finishes.

Model selection

Every candidate ranked, then the champion tuned

No models compared yet
Appears when training finishes.

Deployment

The champion as a live API on Modal

No model yet
Deployment opens up once a run finishes training.

Try the model

Live predictions from the deployed endpoint

No model yet
Deployment opens up once a run finishes training.

Share

A public test page anyone can open, no account needed

No model yet
Deployment opens up once a run finishes training.