> Projects
Selected work in AI, product engineering, and open source.
// featured
LLM-Powered Root Cause Analysis Engine
A Large Language Model integration that automates and accelerates root-cause analysis for silicon test failures. Connects to massive historical test report databases and failure logs, enabling engineers to query years of troubleshooting data using natural language. Instead of manually searching through thousands of reports, the team asks questions like "What caused yield drops on Product X at Fab Y in Q3?" and gets synthesized answers with cited sources. Also automates technical documentation generation from structured test data, cutting report preparation time by over 60%.
Predictive Silicon Health Monitor
A machine learning system that shifts post-silicon validation from reactive troubleshooting to predictive maintenance. Trained on historical test data spanning millions of devices, the model identifies subtle parametric drifts in silicon performance before they escalate into batch failures. Uses gradient-boosted trees and anomaly detection on multi-dimensional test vectors to flag at-risk wafer lots. Integrated with automated alerting so the validation team can intervene early, preventing yield loss and reducing scrap rates. Moved the team from a "wait and fail" model to a "predict and prevent" paradigm.
Silicon Yield Analytics Platform
An end-to-end data analytics platform for monitoring and optimizing semiconductor mass production yields across global fabs and test houses. Ingests streaming test data from UltraFlex systems via KQL pipelines into a relational database. PowerBI dashboards provide real-time visibility into yield trends, bin distributions, and parametric shifts. JMP statistical models identify correlations between process variations and yield loss. Stakeholders across engineering, operations, and management use the platform to make data-driven decisions on test limits, process adjustments, and cost optimization. Reduced yield excursion detection time from days to under 2 hours.
// all projects
Global Test House Performance Dashboard
A real-time PowerBI and SQL-driven dashboard system providing visibility into networking product performance across different outsourced test houses and fabrication facilities worldwide. Aggregates data from multiple test floor systems into a unified view. Tracks key metrics: test time, yield, retest rates, bin distributions, and tester utilization. Enables side-by-side comparison of test house efficiency and identifies outliers. Used by operations teams to make sourcing decisions and by engineering to standardize test quality globally.
Stolbun.com — Digital Identity Platform
This website. A monorepo personal platform built with Next.js 15 and Django 5, featuring a Three.js 3D starfield, glassmorphic design system, and a private family genealogy section gated by JWT middleware. Serves as both a professional portfolio and a secure family data archive with tiered authentication (public, authenticated, family member, admin).
Post-Silicon Validation Automation Suite
A custom Python-based automation framework for post-silicon hardware validation and characterization of high-speed networking chips. Automates test sequences, data collection, and pass/fail analysis across multiple test platforms. Handles bench-level hardware control, instrument communication, and result aggregation. Replaces manual characterization workflows that previously took weeks with automated runs completing in hours. Includes a configuration-driven test plan system so engineers can define new validation sequences without writing code.