Field Notes

Writing on AI, automation, and shipping real things.

What I shipped, how, and what broke. Specificity = credibility.

Tags:
Build LogPart 2AnalyticsAI AgentsNext.jsSelf-HostingSmol Analytics
11 min read
Zyntopia telemetry pipeline connected to self-hosted Smol Analytics and Antigravity AI agent

How I Implemented Smol Analytics for Zyntopia: Self-Hosted Setup and Autonomous AI Monitoring

A complete build log of how I implemented Smol Analytics (created by Arjun Patel) on a private VPS, instrumented Zyntopia's Next.js application, and connected AI agents for autonomous monitoring.

Read article
Build LogPart 11Seepient
14 min read
Abstract editorial illustration of an AI assistant completing a chain of tasks

How I Turned My AI Assistant Into Something That Actually Does the Work

Over six releases my assistant went from a good talker to something that does the work: it refreshes its own model menu, writes checklists with correct numbers, survives a first-run crash, lets you pick how much to trust it, and recovers when a model or a network goes quiet.

Read article
Build LogPart 8Seepient
8 min read
Flat editorial diagram in charcoal and blue on cream: a control room console screen listing model rows with price tags and connected account checkmarks, with four identical doorway icons feeding one shared core.

Every AI Model, One Screen: How I Built Seepient's Control Room

Most AI tools leave you juggling API keys and dashboards while every model behaves differently. I rebuilt Seepient around one control room where every model is visible with its price, sign-in takes seconds, and every doorway behaves the same.

Read article
Build LogPart 1OpenFusionArchitectureOpen Source
9 min read
Minimal editorial diagram in charcoal and brand blue on warm cream showing a multi-model fan-out pipeline synthesizing into a single judge outcome.

Building OpenFusion: How I Built a Local Multi-Model AI Fusion Engine

Why rely on a single AI model when you can synthesize answers across a panel of models? Here is how I built OpenFusion, the architecture behind its two-step judge, and how I solved the MCP client timeout wall.

Read article
Build LogPart 7Seepient
9 min read
Flat editorial diagram in charcoal and blue on cream: a workflow line reroutes over a failed provider node marked with a cross, continuing to healthy provider nodes.

How I Built an AI Assistant That Keeps Working When Its Model Provider Fails

My AI assistant used to depend on a single model provider, so every outage stopped my work. I rebuilt it around a multi-provider engine, and now it keeps working no matter which AI company has a bad day.

Read article
Build LogPart 6SeepientAI StrategyBusiness Automation
7 min read
Strategic business architecture showing one unified AI brain powering multiple company touchpoints

Stop Building Five Different Chatbots: How to Build AI That Actually Runs Your Business

Why most business AI projects turn into expensive maintenance headaches, and how a one-brain approach gives you total control, lower costs, and zero vendor lock-in.

Read article
Build LogPart 5Seepient
7 min read
Cover image for seepient v0 4 3 one screen approvals

Seepient v0.4.3: One-Screen AI Approvals That Survive a Crash

How I rebuilt Seepient's permission UX around a single, complete approval screen, and made every granted permission survive a crash, so your AI assistant's work is never silently undone.

Read article
Build LogAnalyticsAI AgentsNext.jsSelf-HostingOpen Source
7 min read
The AI-Native Analytics Stack with Smol Analytics and MCP integration

The AI-Native Analytics Stack: Why We Switched to Smol Analytics (and How to Set It Up)

Why we moved away from bloated analytics trackers to a self-hosted, cookieless single binary that interfaces directly with AI coding assistants over MCP, along with the complete step-by-step setup guide.

Read article
Build LogZyntopia
6 min read
Diagram of the blog automation pipeline from a GitHub merge to a published article

How I Built an AI Agent That Writes My Blog From My GitHub

The build log of a blog that drafts itself: how merged pull requests become reviewable article drafts, and how one cell in a spreadsheet decides what goes live.

Read article
Build LogPart 2Seepient
4 min read
Automated quality gates and architectural guardrails for AI-generated code

Enforcing Architecture in the Age of AI Code: Taming the Slop

AI development tools speed up development, but unguided code generation introduces hidden operational risks. Here's how I enforce strict governance over AI-built code.

Read article
Build LogPart 3Seepient
5 min read
Seepient four-stage governance pipeline: Policy, Approval, Sandbox, Audit

Inside Seepient's Permission System: Policy, Approval, Execution, Audit

Why simple yes/no approval prompts fail when automating business tasks, and how my 4-stage governance pipeline makes autonomous AI execution safe and fully auditable.

Read article
Build LogPart 1Seepient
5 min read
Six-layer modular architecture diagram for Seepient AI engine

Foundational Architecture: Designing for Evolution, Scale, and Zero Redundancy

How a disciplined, six-layer modular architecture protects Seepient from technical debt, isolates system failures, and guarantees zero vendor lock-in.

Read article
Build LogPart 4Seepient
4 min read
Engineering an AI Person: Seepient persistent digital colleague architecture

Seepient: Engineering an AI Person, Not Another Chatbot

Why single-prompt chatbots fail for busy teams, and how Seepient works as a persistent, secure digital colleague across desktop and cloud environments.

Read article
Build Log
4 min read
Build log: a Slack RAG agent that answers team questions from internal docs

How I Built a Slack Agent That Answers Team Questions From Internal Docs

A walkthrough of the RAG assistant I built for an ops-heavy team: what it does, the stack, where it nearly broke, and what I'd do differently. It's concrete, not conceptual.

Read article