Visibility, Replay & Trust

How I’ve made ambiguous systems usable — before sonar
What I’ve Built and Owned
  • Interpretability dashboards for ML-driven systems
  • Trend analysis over months and years
  • Side-by-side comparison of historical vs current results
  • Human-facing summaries layered on raw data
  • Tools designed for operators, not model builders
TesTU — Interpretability in Practice
TesTU dashboards · performance trends · regression views
  • Time-series telemetry from real users
  • Explaining why predictions changed over time
  • Detecting drift vs genuine behavioral change
  • Making ML results legible to non-ML stakeholders
Why This Transfers Directly to Sonar
  • People interpreting data they didn’t capture
  • Delayed analysis and cross-team handoff
  • Need to compare “now” vs “then” safely
  • Trust depends on visibility, not accuracy claims
Acoustic Dashboards I Built
Spectrograms · window summaries · anomaly rates · timelines
  • Spectrograms and window-level summaries
  • RMS, centroid, anomaly-rate trends
  • Artifacts designed for comparison and replay
  • Focus on explaining behavior, not asserting labels
My value is not producing answers — it’s building systems people can question, trust, and improve over time.