Back to engineering archive
AI & Speech
Bastian Data Studio
Interactive web studio for tokenizing and preparing AI training datasets
A modern developer playground built with Next.js 15, Zustand, and Tailwind CSS for cleaning, normalizing, and formatting raw text datasets before fine-tuning LLMs.
CategoryAI & Speech
LinksPrivate
StackNext.js 15, TypeScript
ScaleNext.js 15
The Challenge
Pre-training and fine-tuning custom machine learning models requires extensive data scrubbing—removing special characters, standardizing encodings, and splitting training/validation splits.
Architecture & Solution
Built an interactive client-side playground with fast live previews, token counters, regex transformer presets, and JSONL format exporters.
Engineering Decisions
- Next.js 15 + Zustand Reactive StatePaired React 19 Server Components with a lightweight client-side Zustand store for instant, zero-lag dataset transformations.
Outcomes
- Accelerates dataset preparation workflows for natural language processing experiments