Browser-local AutoML with ONNX-oriented export packages

What browser-based AutoML means

Traditional AutoML usually runs in a notebook, a desktop environment, or a hosted platform. Browser-based AutoML moves much of the interactive workflow into the browser. For MLdeck, that includes CSV profiling, target selection, preprocessing configuration, baseline comparison, model training, evidence review, and ONNX-oriented export package preparation for tabular classification and regression tasks. The goal is not to replace every part of a mature machine learning platform. The goal is to make the first modeling loop faster, easier to share, and less dependent on raw data transfer.

Browser-based describes the user experience: the workflow is available in a web browser. Browser-local describes the execution boundary for normal MLdeck CSV training: raw CSV rows are processed in the browser rather than uploaded to a cloud training job. The terms overlap, but they answer different search intents.

This matters because many CSV modeling questions are exploratory. A business analyst wants to know whether churn labels contain signal. A student wants to understand leakage. A founder wants to evaluate an early dataset. A data scientist wants a quick baseline before writing a deeper pipeline. Browser-based AutoML gives those users a lower-friction place to begin.

How WebAssembly enables local ML workflows

Modern browsers can run WebAssembly, a low-level execution format designed for portable, sandboxed applications on the web. MLdeck uses WebAssembly-based Python tooling so familiar machine learning components can run in the browser environment. That makes it possible to use scikit-learn style workflows without requiring a local Python installation, virtual environment, package manager, or notebook server.

WebAssembly does not make browser training magic. It still uses the user's device. CPU speed, memory, browser limits, file size, feature count, and algorithm choice all matter. For browser-local CSV modeling, this tradeoff is often worthwhile. You avoid setup, reduce raw-data movement during normal training flows, and get fast feedback on whether a dataset deserves deeper validation.

From CSV upload to model comparison

The MLdeck workflow begins with a CSV file. The app profiles columns, estimates types, shows missingness and cardinality signals, and helps the user decide which column should be predicted. From there, the user can review feature inclusion, choose or confirm a target, and train candidate models. The leaderboard and warnings are designed to support interpretation rather than hide the details behind a single score.

Browser-based AutoML is especially helpful when the question is, "What should I try next?" It gives a structured path from raw table to model evidence, but it still encourages users to think critically about ID columns, leakage, class imbalance, changing date ranges, and whether a target is actually available at prediction time. MLdeck's key technical differentiator is that it can generate ONNX-oriented export packages directly from browser-local training workflows, allowing users to test and validate portable model artifacts without uploading raw CSV data during normal browser training flows.

Why browser execution changes privacy and setup requirements

When training runs browser-local during normal flows, raw CSV data does not need to be sent to a cloud training job just to review it. That can be useful for internal datasets, classroom data, practical CSV modeling, or privacy-sensitive workflows. Browser execution also reduces setup: no Python install, no dependency resolution, no notebook server, and no cloud account just to test a basic tabular model. One network dependency exists at training start: the browser ML runtime may be downloaded from a third-party CDN. This is a code download — raw CSV rows and file bytes are not sent to it. See the privacy-first page for details.

Backend services may still exist for app, account, support, and control-plane features. Users are also responsible for local device security, browser extensions, downloaded artifacts, and any external systems used after export. Browser-based AutoML reduces friction; it does not remove the need for responsible data handling.

Local Advisory Playbooks and optional AI guidance

MLdeck can generate a local Advisory Playbook from run metadata in the browser. Optional AI Copilot guidance and Optional AI Advisory Notes can explain metrics, validation limits, feature importance, export readiness, and next experiments, but only after session-scoped metadata-sharing consent. Optional AI features use sanitized metadata such as column names, target name, model names, metrics, preprocessing steps, validation/export status, and the user's question. They do not inspect raw CSV rows, uploaded file contents, row samples, model binaries, package binaries, prediction curves, or large artifacts.

Where browser-based AutoML fits

MLdeck is a live browser-local AutoML product for practical CSV-based machine-learning workflows. It helps users move from CSV profiling to model comparison, warnings, reports, and export packages while keeping the normal training workflow local to the browser.

It is also worth separating convenience from evidence quality. A browser workflow can make training easier to start, while MLdeck helps users inspect sampling, target leakage, class imbalance, temporal drift, and whether the future prediction setting matches the CSV. Start with Data Quality for Machine Learning, then review AutoML validation evidence and export artifact boundaries before relying on results.

Browser-based AutoML FAQ