What Building My Own AI Bot Taught Me About Generative AI
I built a bot trained on my own X bookmarks and likes. Around 50,000 of them, accumulated over years...
I build AI agents and SEO automation tools that actually work — no hype, no fragile demos. Python, Playwright, Claude, Cloudflare.
"I did everything the AI era asked. Learned the frameworks, built the projects, wrote the threads. It still didn't pay my bills. So I stopped chasing hype and started shipping tools that actually solve problems."
Production-grade agents that browse, process data, write and audit content, and integrate with your tools. Built to run unsupervised, with resumable pipelines so nothing is lost on failure.
Real-browser SEO audits, Google Search Console analysis, backlink qualification, and content gap tools. I run actual Playwright browsers — so the output matches what Google actually sees.
Workers, Pages, D1, KV, Vectorize, and edge deployments. Fast, globally distributed, and cheap to run. My default infrastructure stack for a reason.
Edge-native MCP servers built on Cloudflare Workers — hybrid vector search, cross-encoder reranking, and LLM reflection layers. No external infrastructure. Live example: 100k+ documents at $5/month.
Scientific computing (scipy, matplotlib, pandas), AI agents, browser automation, and serverless APIs. Remote. Production code with published writeups for every project.
White-label automations, Claude workflow consulting, and production pipelines for teams that want to add AI capabilities without building from scratch. I bring the architecture, you keep the client.
Not sure which fits? Email me to scope it — I'll tell you honestly if I can help and what it would take.
My production RAG pipeline had 22 days before its reflection model was deprecated. Cloudflare's recommended replacement: Gemma 4 MoE. I migrated and benchmarked every step. 728 reflections generated. $5/month cost unchanged.
An MCP server that turns any Claude conversation into a knowledge base. Embed, store, and query vectors on Cloudflare Vectorize — no external infrastructure. Live dashboard at vectorize-mcp-worker.fpl-test.workers.dev/dashboard.
The formula can be perfect. The model can be correct. Yet the result is still dry — because the assumption about the environment was wrong.
Real-browser SEO audit using Python + Playwright + Claude. No API scraping. Real browser, real data, real results.
Backlink analysis at scale gets expensive fast. I reversed the problem and cut per-URL costs to near zero.
A local AI agent that reads my dev.to posts, scores them against SEO and readability heuristics — zero cloud costs.
Close the Click Gap — upload your Google Search Console export and get specific title & meta rewrites that turn impressions into clicks. Built with the same GSC + Claude pipeline I use on client sites.
I built a bot trained on my own X bookmarks and likes. Around 50,000 of them, accumulated over years...
Most agent memory systems are digital attics. You put things in. You hope to find them later. You...
This is a submission for the Hermes Agent Challenge For the past six months, I have been building...
This is a submission for the Google I/O Writing Challenge I was already running MCP servers on my...
I built a SERP feature detection module for my SEO agent. Then I ran it on the queries I'm targeting...
Email me with what you're trying to build. I'll tell you honestly what's feasible, how long, and what it costs. No sales deck. No follow-up sequence.
Email Me