AutoData

Raw datasets in, ML-ready data out.

pipeline stages at handover: 6
native data connectors: 30+
downtime in the 31 GB → 992 GB cutover: 0

Built and operated at BGTS · Dec 2025 – Apr 2026

AutoData is a full-stack data-preparation platform for machine learning: connect a source, choose a pipeline, and get back a dataset a model can train on.

I built and operated it at BGTS, and migrated it live from a 31 GB machine to a 48-core / 992 GB VM without taking it down.

What is in it

01A pipeline that runs from LLM-assisted data completion and cleaning, through imputation and scaling, to synthetic data generation.
0230+ native connectors across databases, warehouses, storage, streaming and SaaS — Snowflake, BigQuery, Databricks, S3, MongoDB and Kafka among them.
03Published Python SDK on PyPI, backed by a REST API.
04Zero-downtime production migration via live public-IP cutover to a 48-core / 992 GB VM, eliminating OOM failures.
05Security hardening: LLM prompt-injection guards, CSRF protection, Fernet credential encryption, read-only SQL validation.

Screens

datatoolpack.com
autodata.datatoolpack.com