Refract AI/ML Platform Documentation

The Product

Refract (by Fosfor) is an enterprise AI platform that brings together AI frameworks and templates to prepare, build, train, and deploy machine learning models. It reduces up to 70% of the effort for data science teams by automating pre- and post-model development steps — data provisioning, preparation, model deployment, governance, and monitoring.

This is the kind of product where documentation isn't optional — users need clear guidance to navigate complex ML workflows, understand automation features, and configure pipelines correctly.

The Problem

When I joined, the product was in active development with significant gaps in documentation:

Existing docs didn't reflect recent changes, new features, or deprecated functionality
The user guide was maintained as a Word file — no version control, no searchability, no scalable publishing
Documentation lacked accessibility features and didn't align with current design guidelines
A full product rebrand (to Fosfor) was on the horizon, requiring coordinated content updates

What I Built

I owned the documentation from the ground up, starting with an audit of existing content and a migration plan:

Migrated from Word to Sphinx — moved the entire documentation system to Sphinx (Python-based docs-as-code generator), enabling version control, automated builds, and structured publishing
Built 100+ topic online help — scoped, planned, and wrote a comprehensive help system covering all platform features, What's New content, and release notes
Created how-to video series — produced step-by-step video tutorials for complex workflows, improving user comprehension of automated ML features
Managed the rebrand transition — updated all visual elements, terminology, tone, and brand identity across the full documentation set when the product rebranded to Fosfor
Aligned with design guidelines — worked with the design team to standardize formatting, visual elements, and information hierarchy across all docs

Impact

Reduced support requests by giving users self-service access to up-to-date, searchable documentation for all platform features and ML workflows.

Video tutorials provided visual guidance for complex AI workflows, improving adoption among users new to automated ML pipelines.

Docs-as-code migration (Word → Sphinx) established a scalable, version-controlled documentation system that the team could maintain as the product evolved.

Why This Matters for What I Do Now

This was the project where I first documented AI/ML systems at production scale — understanding model workflows, data pipelines, and governance requirements from the inside. The skills I built here (information architecture for complex AI platforms, docs-as-code with Python tooling, translating ML concepts for diverse audiences) are the same ones I use today building RAG systems and developer content.