Technical Focus
Agentic AI infrastructure, security architecture, and production migration methodology.
100+ companies. 35 years. Every paradigm shift from assembly to agentic AI — and the production scars to prove it.
Agentic AI & Claude Code
Building production-ready multi-agent systems across Claude Code and GitHub Copilot: specialized agents, deterministic control via hooks, 140x token-efficient skills, self-improving knowledge harvest, and team architectures that scale beyond single-context limits.
Claude Code Series
View all →Multi-Agent Frameworks
Combining the Power of LangGraph with Pydantic AI Agents 2025 → AI Agent Blueprints: Implementing Anthropic's Framework with Pydantic AI 2025 → Building Domain-Specific AI Agents with LangGraph and Pydantic AI 2025 → Advanced Testing Strategies for LangGraph and Pydantic AI Agent Systems 2025 → Scaling LangGraph and Pydantic AI Systems: From Prototype to Production 2025 →Claude Code Bootstrap Framework
An agent swarm that builds agent swarms. A 12-step pipeline where Claude Code agents analyze any codebase and generate complete Claude Code infrastructure -- agent teams, hooks, skills, and slash commands -- in 30-55 minutes. Three production migrations validated. The second was harder but faster.
AI Security Architecture
Defense-in-depth security for AI agent systems: OWASP Top 10 for Agentic Applications coverage, multi-tier trajectory monitoring, input sanitization patterns, and per-archetype security configurations across 7 project types.
Production AI Systems
Three completed AI systems proving end-to-end methodology: natural language to SQL dashboards, knowledge graph pipelines for 1,000+ research notes, and autonomous job market intelligence across 1,975 companies.
GitHub Copilot Agent Pipelines
FeaturedSeven specialized Copilot agents that form a structured development workflow for a legacy Servoy enterprise application with 10,000+ functions, 1,000+ files, and 22 modules. Neo4j graph-powered code intelligence, cross-model orchestration, and a self-improving knowledge loop where every code review makes every agent smarter.
Adversarial Agent Testing
FeaturedAI agents that attack each other to find vulnerabilities. Red Team probes, Blue Team defends, a Referee scores both -- all using Claude Code with worktree isolation. Two rounds of live exercises against a real target drove ASR from 65% CRITICAL to 47% HIGH, with a regression wave proving patches hold at 20% and an escalation wave exposing architectural gaps at 85.7%.
Obsidian Notes Pipeline: AI-Powered Knowledge Management
FeaturedA full-stack RAG application that transforms YouTube videos into interconnected Obsidian notes -- 1,000+ notes, 2,757 auto-generated links, 5,000 searchable chunks, and a chatbot with 2.5s latency, all for $1.50.
Data Intelligence & GraphRAG
From vector embeddings to knowledge graphs to GraphRAG: building the retrieval infrastructure that grounds AI in real-world knowledge. Includes a 4-part production optimization series with benchmarks.
Foundations
Vector Databases: The Engine Powering Modern AI Applications 2025 → RAG: Grounding AI with Real-World Knowledge 2025 → Graph Databases: The Foundation Enabling Context-Aware AI Applications 2025 → GraphRAG: Enhancing Retrieval with Knowledge Graph Intelligence 2025 →GraphRAG Deep Dive (4-Part Series)
GraphRAG: Building Bridges in the Knowledge Landscape - 1 of 4 2025 → Optimizing GraphRAG: Five Essential Techniques for Production Performance - 2 of 4 2025 → Optimizing Parallel Relationship Loading in Graph Databases: The Mix and Batch Technique - 3 of 4 2025 → Benchmarking and Optimizing GraphRAG Systems: Performance Insights from Production - 4 of 4 2025 →Technical Proficiency
Research Directions
Deterministic Control via Hook Engineering
A CLAUDE.md instruction achieves ~90% compliance. A hook achieves 100%. Per-agent hook embedding scales better than global hooks — and every gap found across 3 migrations occurred in an area without hook enforcement.
Read the article →Progressive Disclosure Skills Architecture
Three-tier skill loading makes domain knowledge accessible without the token cost of loading everything simultaneously. Current exploration: how skill content should evolve as projects mature, when to split vs. merge, and how stale content gets detected.
Read the article →Agent Prompt Injection Defense
XML boundary delimiters wrap external input so agents distinguish instructions from user content. A PostToolUse hook scans every file read for 22 injection patterns across 10 categories — role-play injection, instruction override, base64 payloads, hidden HTML comments — flagging suspicious content before it reaches agent reasoning.
Read the article →Agent Trajectory Monitoring
Most Claude Code projects monitor individual tool calls — nothing monitors the pattern over time. A 3-tier system: heartbeat checkpoint (every 25 calls, 5 anomaly patterns), orchestrator watchdog timer, and optional Haiku-based trajectory analysis.
Agentic AI Security
Defense-in-depth for AI agent systems: OWASP Top 10 for Agentic Applications coverage, multi-tier trajectory monitoring, per-archetype security configurations across 7 project types, and rate limiting as circuit breakers. Every agent must prove who it is, justify what it wants, and earn trust continuously.
Read the article →GitHub Copilot Agent Pipelines
10 specialized Copilot agents across 4 workflow types: development (research → architect → developer → reviewer), domain capture with resumable frontier exploration, self-improving skills via knowledge harvest, and JIRA-integrated handoff documentation. The same team architecture principles proven in Claude Code, applied to a different ecosystem.
Building Agentic AI Business Solutions
Taking real-world workflows and building advanced agentic AI solutions: natural language to SQL dashboards, knowledge graph pipelines for research notes, autonomous job market intelligence. Each project proves the methodology end-to-end, from messy data to production system.