Project Phenotype

Problem

Research teams were tracking model lineage in disconnected tools, making it difficult to explain why specific outputs emerged and how architecture changes influenced behavior over time.

Role

Led systems architecture and interaction design, with direct ownership of data modeling, query strategy, and storytelling layer for research reviews.

Process

1. Defined a graph-first schema for lineage events, mutation metadata, and generation context.
2. Built a timeline + cluster visualization that keeps macro trends and micro events readable at once.
3. Introduced a curation workflow so each release can be narrated with evidence, not screenshots.
4. Validated the model with weekly research walkthroughs and adjusted affordances for non-technical stakeholders.

Results

• Reduced investigation time for anomalous generations from hours to minutes in internal reviews.
• Created a reusable report format that aligned engineers and researchers around the same evidence.
• Improved confidence in experiment tracking quality before publication handoffs.

Reflection

The key lesson was that visualization alone is not enough: a durable representation must also support narrative sequencing. We now design data views and explanation flow as one system.