01
Chat-first agent
Describe the task in plain English or Arabic. The Gemini-powered brain interprets it, inspects your dataset, and proposes a plan you can approve.
▎AutoML, with words.
Upload a dataset. Tell the agent your goal. Get a trained, evaluated, deployable model in minutes — no notebooks, no boilerplate.
/ 01 — capabilities
Describe the task in plain English or Arabic. The Gemini-powered brain interprets it, inspects your dataset, and proposes a plan you can approve.
Cleaning, encoding, scaling, model selection, cross-validation, hyperparameter tuning — handled. You get the winning pipeline, persisted and downloadable.
CSV, Excel, parquet for ML. Images, text, audio routed to PyTorch / TensorFlow with mixed-precision training on Kaggle T4 / P100 GPUs.
Datasets larger than memory? Chunked, parallel processing keeps the workflow identical from 1 MB to 100 GB.
Every trained model gets an instant prediction interface — type in new feature values, see the answer and class probabilities in real time.
Export your model as a joblib bundle. The whole pipeline is portable to any Python environment.
/ 02 — how it works
OAuth 2.0 — no password to remember. Free tokens credited automatically.
Drop a CSV / Excel file or paste a Drive link. The agent profiles columns, missing values, and suggests a target.
“Predict customer churn.” The agent picks classification, splits data, trains 4 candidates, and reports the best one.
Test predictions in the live playground, download the artifact, or hand it to your API.
/ 03 — live demo
Sign in, drop a CSV, and ask the agent to build a model. You'll see column profiling, automatic task detection, cross-validation across candidate models, and the winner saved to your dashboard — all in under a minute on a typical tabular dataset.
Open the playground →/ 04 — early users
“I described my churn problem in two sentences. Twenty seconds later I had a model with 0.91 AUC. This is what AutoML should always have felt like.”
“The chat interface removed the ‘which library do I install today’ question. My grad students just build.”
“We pushed a 2 GB CSV through and it shrugged. Same workflow, same UI. That was the moment we paid for Pro.”
/ 05 — faq
No. The platform is designed for non-coders. If you do know Python, you can also download the trained pipeline and integrate it anywhere.
On the Free plan, 200 MB per upload. Pro raises the limit and routes larger files through Dask streaming. We've tested up to ~10 GB on a single Pro project.
You get 1,000 free tokens on signup. Training consumes 50 tokens; predictions 1 token each. When you run out, upgrade to Pro or Enterprise — no second-account loopholes.
Yes. Every trained pipeline is downloadable as a joblib artifact that includes the preprocessing steps. It runs in any Python 3.11+ environment.
Datasets are stored under your account and never used to train shared models. Delete a project and the underlying files are removed too.