Work & Architecture
How do we make software faster, understand AI models, and coordinate systems that can disagree? Explore the code, experiments, and trade-offs behind each project.
Helping PyTorch find an optimization it could miss
See how making a small calculation visible gives the compiler more opportunities to combine steps.
Explore the visual explanation →Case studyTeaching Apple's converter two missing image operations
See how two merged contributions translate unsupported image operations into Core ML building blocks.
Explore the visual explanation →Aegis
AI-powered autonomous security scanner for GKE microservices. Automatically discovers, analyzes, and secures 11 microservices with AI-generated security summaries.
K2-Consensus
Real-time reasoning team implementing distributed consensus algorithms for multi-agent coordination and decision-making.
AzureDev Nexus
Intelligent developer ecosystem in .NET with AI-powered code reviewer using Azure OpenAI and real-time collaboration via SignalR.
Graph-Aware Anomaly Detection
Engineered a high-throughput anomaly pipeline for Trufflow: XGBoost at 1.2M events/sec and logistic regression at 9.8M events/sec with 87.5% recall one hour ahead.
MedInsight
GKE-based polyglot microservices architecture (Go/Python) with Gemma 2 reasoning. 89%+ automated test coverage, HIPAA-compliant medical diagnostics platform.
LLM Fine-Tuning with GRPO
Fine-tuned Microsoft Phi-4 (14B) using Grouped Relative Policy Optimization for GSM8K math reasoning. Memory-efficient LoRA training with Unsloth.
AuraFlow
Cloud-native Java Spring Boot microservices architecture on AWS, processing 40,000 events/min with event-driven design and Saga pattern.
SecondSight
Agentic accessibility engineer that audits live websites with Playwright and Gemini, maps WCAG failures to source files, and proposes pull-request-ready fixes.
GKE Security Agent
Built for GKE Turns 10 Hackathon. Automated security monitoring and threat detection for Google Kubernetes Engine clusters.
Project Genesis: How a Software Factory Decides to Try Again
Follow a simulated build through planning, deployment status, cost review, and revision. Learn why an agent loop needs explicit state, evidence, and stopping rules.
K2-Consensus: What Does an AI Panel's Agreement Actually Mean?
Explore a proposed expert-panel workflow through three critiques and a moderator. Learn how useful disagreement differs from voting, evidence, and distributed consensus.
Trufflow: How Would You Know an Early Warning Works?
Follow an alert backward through thresholds, time windows, and evaluation splits. Learn why a strong aggregate score can hide weak service-level behavior.
AzureDevNexus: Designing the Review Experience Before Connecting the Model
Walk through a .NET developer-tool prototype to understand review contracts, sample data, persistent history, and the difference between a live update and a saved result.
Trufflow: A Million Events per Second Can Still Mean a Waiting User
An illustrated guide to batching, preparation cost, and model choice—and why throughput, amortized inference time, and alert latency need different measurements.
Testing MedInsight: What Happens When a Model Worker Fails?
A companion to MedInsight's architecture walkthrough: use duplicate jobs, worker crashes, and incompatible messages to reason about reliability across Go and Python.
Trufflow: Why an Anomaly Needs Its Neighbors
Follow a latency spike through Trufflow's design to understand service-specific baselines, graph context, and the gap between an unusual event and a useful alert.