· 7 min read
Semantic Entity Resolution Without LLM: When Cosine > 0.92 Is Enough
How a three-level resolver cuts entity deduplication costs by 85% by reserving LLM calls for the 5% of cases that actually need them.
> I build AI products solo.
> 24 products. 0 employees. $15/mo infra.
I build products with AI agents. Not the "AI-powered" marketing kind — actual autonomous agents that write code, run tests, deploy to production, and fix bugs while I sleep.
My setup is called Factory OS. It's an orchestration layer on top of Claude Code with 15 specialized agent roles: Builder, Discovery, Strategy, Quality, DevOps, Marketing, and more. The CEO agent delegates tasks, the Builder agent writes code, the Quality agent reviews it. I approve commits.
24 products shipped (8 standalone + 16 systems inside AICPO Platform). Total infrastructure cost: $15 per month. A used NVIDIA P100 graphics card for $150 handles the ML inference. Everything else runs on free tiers — Cloudflare, Vercel, Supabase, Groq.
The future isn't about writing code faster. It's about not writing code at all. It's about describing what you want and watching agents build it. That future is here. I'm living it.
## Products
Tool Sites
LIVE199 programmatic SEO pages
Astro Tailwind MDX
199 pages across 5 domains
OpenClaw Finance
BUILTAI finance agent factory
Python DeepSeek R1 Docker
RackCloud
BUILTCustom cloud platform from 400+ servers
Claude Opus Proxmox Python
## Latest
· 10 min read
PostgreSQL as a Graph Database: CTE + pgvector + pg_trgm = No Neo4j Needed
How to build a production knowledge graph on PostgreSQL using recursive CTEs, pgvector, and RRF fusion — without touching Neo4j.
· 9 min read
Bi-Temporal Knowledge Graph: Why Facts Get Invalidated, Not Deleted
Why overwriting facts destroys institutional memory — and how bi-temporal storage preserves the full history of what you knew and when you knew it.
## Newsletter
Follow the build
One email when I ship something new. No spam. Unsubscribe anytime.