The work, up close.
The products, the decisions behind them, and the engineering that makes them work.
Codelit
From a conversation to a working system.
Product plans, architecture, and supervised agent work, connected in one Thread.
Read the case study
ResolveMesh
Find the capability. Show the evidence.
Explainable tool discovery for AI agents, with honest confidence and visible ranking.
Read the case study
AbleMakers
A little support. A lot of possibility.
Free, accessible AI learning for people with intellectual disabilities and the people supporting them.
Build storyInside the engineering
Focused case files on architecture, interfaces, infrastructure, and AI systems.
ResolveMesh: Evidence-First Agent Tool Discovery
How I built and benchmarked an explainable capability resolver for agents without invoking a single discovered tool.
AI Architecture Generation
How I turn a plain-English system description into an editable architecture, while keeping model output separate from validated application state.
Interactive Architecture Canvas
How I built Codelit's editable architecture surface around stable graph identity, readable layout, recoverable edits, and inspectable component context.
GitHub Repository Analysis
How I turn sampled repository evidence into an editable architecture hypothesis, with explicit limits on coverage, inference, and GitHub permissions.
Multi-Provider AI with Explicit Failure Boundaries
How I separate provider routing, response validation, and recovery in Codelit, without treating a fallback model as an uptime guarantee.
Infrastructure Export
Turning an architecture into infrastructure scaffolding, with the configuration, security, and deployment decisions that generated files must leave explicit.
AI Architecture Review
Using an AI architecture review to surface questions and missing assumptions, without mistaking a diagram for a security audit or capacity test.
Architecture Templates with Visible Assumptions
A curated architecture library for exploration and design discussions, with a clear distinction between reference patterns and verified production systems.
System Design Learning Through Editable Models
How I connect architecture explanations with editable models, so learning moves from recognizing a pattern to defending a specific design decision.