Building Arc: A Self-Learning ERP Agent That Audits Clients Before Writing a Single Line
Phase-gated autonomous agent with a SQLite knowledge base, infrastructure scanner, and human-in-the-loop approval gates before any code is generated.
// The diary
All articles
16 posts on agentic systems, ERP automation, Claude Code, and building products solo.
Building AI Agent Systems in 2026: Multi-Agent Orchestration, Memory, and Tool Calling
Build AI agent systems in 2026. Multi-agent orchestration, SQLite vector memory, tool calling, and state management for production agent architectures.
Replacing NLP Pipelines with LLM Agents: MIS v2 Architecture
Why I scrapped spaCy and Hugging Face for a Batch-and-Specialize pattern using Gemini's 1M context window — 40% faster, half the code.
Personal Claude OS: ~/.claude/ as a Portable Developer System
Slash commands, SOUL.md agent personas, lifecycle hooks, and permission gates — as a deployable, version-controlled package.
The Technical Founder's Prompt Stack — Gumroad Launch
50 production-grade prompts for ERP consulting, agentic pipelines, and AI content operations. Numbers from the first 30 days.
Claude Code Complete Mastery Guide: The Tool You Are Not Using Correctly
Claude Code is the most powerful AI coding tool in 2026. Master slash commands, MCP servers, approval gates, and agentic loops to ship 10x faster.
I Stopped Prompting Claude and Started Operating It
Personal Claude OS architecture — identity layer, memory management, hooks, skills, agents, and the self-improvement loop. How I turned ~/.claude/ into a portable operating environment.
DAG vs Sequential Agent Execution: Choosing the Right Orchestration Pattern
DAG vs sequential agent execution comparison. When to use DAG-based orchestration vs sequential pipelines for multi-agent systems.
OpenClaw 2026 Setup Tutorial: From Zero to Your First Autonomous Agent
OpenClaw 2026 setup tutorial. Install, configure, and run your first autonomous agent with OpenClaw in under 30 minutes.
Build Your AI Chief of Staff: A Complete Claude Code Personal OS Guide
A step-by-step blueprint for building a localized, self-improving cognitive workspace with Claude Code — from environment setup and MCP connections to auto-dreaming and parallel sessions.
MCP Server Configuration: The Production Patterns I Actually Run
Production MCP server configuration for Claude Code. 7 servers, one .mcp.json, context-aware chaining, and the decision tree for adding or removing each one.
Build Your Own LifeOS: A Complete Life-Admin Operating System
A complete spec for building lifeOS — a local, owned, personal life-admin system with four tabs, nine AI skills, and zero vendor lock-in.
Browser-Native Micro-SaaS: 6 Product Ideas That Need Zero Backend Infrastructure
6 profitable browser-native micro-SaaS product ideas using WebGPU, WASM, and on-device AI — with zero server infrastructure and no data egress.
DOCR: I Built a React App That Extracts Structured JSON from Invoices Using Claude Vision
How I built a browser-native invoice OCR app using React 18, Claude Vision, and PDF.js — with a 5-step extraction pipeline, three result views, and CSV/JSON export.
Real-World Multi-Agent Workflows That Ship to Production
How production multi-agent systems actually work: orchestration, state, verification gates, and human approval.
My B2B Outreach Agent Marked a Catch-All Domain "Verified" — Here's the Schema Bug
Building a waterfall enrichment pipeline for B2B lead gen exposed a gap: my CSV schema had no field for the one thing my error-handling policy depended on.