Production

Analyze

Jalaran Analyze is a guided machine-learning workspace that takes a spreadsheet and returns trained models, metrics and plain-language findings without you writing code.

Open Analyze →Create free account

Who Analyze is for

For analysts and engineers who can read a confusion matrix but do not want to spend an afternoon assembling a pipeline to produce one. It is also for the person who has a dataset and a question and genuinely does not know which algorithm the question implies — because that choice is the part Analyze makes for you.

What Analyze does

You upload a file and Analyze profiles it first, telling you what it found before anything is fitted. Then you pick a question rather than an algorithm — predict a number, predict a category, group similar records, find the odd ones out, or explain what drives an outcome — and it runs real scikit-learn pipelines behind that choice: imputation, scaling, one-hot encoding, a cross-validated model leaderboard and hold-out metrics. Heavy fits run on a separate worker process, so a thirty-minute job never blocks the interface, and a written narrative explains the result in prose alongside the charts.

Five workflows: regression, classification, clustering, anomaly detection, explainability
Cross-validated leaderboard with hold-out metrics and feature importance
Saved models re-score new files; PDF export of the full analysis
Uploads and models auto-purge on a retention schedule

How Analyze works

Upload and profile
CSV, JSON or Excel. Analyze profiles the columns and shows you what it found before fitting anything.
Choose a question, not a model
Regression, classification, clustering, anomaly detection or explainability. The algorithm choice follows from the question.
Let the worker run it
Jobs execute on a dedicated worker, not in the web request, so a long fit cannot hang the page. Progress is polled live.
Read the leaderboard and the narrative
Cross-validated model comparison, hold-out metrics, feature importance, and a written explanation of what the numbers mean.
Save a model or export
Refit the winner on the full dataset and reuse it to score new files, or export the analysis as a PDF.

What Analyze does not do

Analyze covers classical machine learning — the scikit-learn family — and deliberately stops there. There is no deep learning, no neural network training, no time-series forecasting and no computer vision. On large files it fits on a representative sample rather than every row, and it says so in the results instead of pretending otherwise. Uploaded data and saved models are deleted automatically after the retention window, so it is a workspace, not an archive.

Common questions

Do I need to know which algorithm to use?

No. You choose a question in plain language — predict a number, group similar records, find outliers — and Analyze selects and compares the appropriate model family, showing you the leaderboard it used to decide.

What happens to my uploaded data?

It is stored in a private bucket that is write-only to the service key, and both the file and any saved model are purged automatically once the retention window passes. Nothing is kept indefinitely.

Can I reuse a model on new data later?

Yes. Saving a model refits the winning pipeline on your full dataset; you can then upload a new file and score it. Saved models are integrity-signed, and a tampered or unsigned model is refused rather than loaded.