Home

IBM InstructLab

Synthetic Data Review Interface for WatsonX

Timeline: Aug 2024 to Jan 2025 Role: AI Technical Consultant Team: 4 designers
IBM InstructLab

Overview

Replaced ad-hoc spreadsheet and CLI workflows with a Figma-prototyped interface that cut review time ~3x across 10,000+ Q&A pairs. IBM SWE Jacob Engelbrecht called it "a big step forward, a large improvement." Selected for internal implementation into WatsonX.

~3xreview time reduction vs. spreadsheet workflows
10,000+Q&A pairs standardized in testing

How it worked

InstructLab trains enterprise LLMs with synthetically generated data, but the existing flow asked reviewers to move data into CSVs or use a CLI. Quote from a developer in interviews: "If I'm unsure about the process or answer, I just hope that someone gets around to reviewing the question." I designed a two-view interface that kept the developers' CLI mental model in the list view while layering a richer review experience on top.

  • List view. Spreadsheet-like, fast scanning, batch approve / deny.
  • Modular view. Expanded per-question review with embedded reference documents, editing, and detailed comments.
  • Collaboration tools. Tagging, commenting, filtering across reviewers; "To Review" vs. "Reviewed" status tabs.
  • Embedded reference docs. Killed Ctrl-F context switching, so reviewers stay grounded in source material without leaving the page.
  • Stack: Figma (prototyping, design), Zoom (user testing with ~5 IBM developers).

Recognition

"This is a big step forward from what we've been doing in the past, a large improvement."

Backend Software Engineer, IBM

Project Documents

Case Study

Stakeholder Presentation