Open to AI-safety and ML roles Hyderabad, India --:-- IST

Mohammed Faisal Parvez

AI Safety & Machine Learning Engineer

Builds systems that decide when a signal is trustworthy enough to act on. Model auditing and interpretability, benchmarks that publish their failures, and bounded-autonomy controllers from GPU processes to accounts payable.

PROBE ACTIVE
0 1 1 false positive rate true positive rate
Decision Boundary Threshold (τ) 0.50
True Positive100%
False Positive100%
DiagnosticDRIFT
AUROC 1.00 — probe separated fine-tuned model from base perfectly. It was detecting fine-tuning drift, not a secret loyalty. The points behind it collapse onto this curve as you scroll.

Lab · Manifold

9,216 points (96 × 96) on a surface that breathes; the points near the zero crossing, the decision boundary, glow in the signal colour. The home hero's field, in depth: as the hero scrolls away they collapse onto the ROC curve.

Source and licence
written for this site on Three.js 0.186, bundled by esbuild. site author · MIT (three). Listed in LICENCES.md.
JavaScript on this page
133.3 KB gzipped for the effect, 201.3 KB in total. The home page carries 68.0 KB.
Reacts to
scrolling collapses the points onto the ROC curve, as the home field does; the pointer tilts the camera and lifts the surface; time.
Off-screen, hidden tab, reduced motion
Paused, paused, one settled frame.