Real systems I've built — AI evals, automation agents, and micro-SaaS — with the honest numbers behind them, plus build-in-public notes.
A tour of how AI became my default toolkit as a product manager: research through Confluence MCP, competitor teardowns via a browsing agent, self-serve analytics on Superset and ClickHouse, quick dashboard and user-story drafts, and even a scoring tool for a personal decision — one workflow, many jobs, every output validated by hand.
A full SMM automation suite for Pixovo — AI-written posts for X, Instagram, Threads and Facebook, scheduled and published on autopilot. 117 posts later: zero organic signups. Here's the system, the numbers, and why it failed.
Instead of shipping AI agents that do tasks, I built one agent that builds automations: you describe a recurring chore in Slack, a Claude-powered agent assembles an n8n workflow, and the workflow runs autonomously from then on. Six-layer architecture, RBAC and audit trails, markdown knowledge with no vector DB — and a cost model of $5–10 to build a workflow, ~$0 to run it.
How we measure search quality on a classifieds marketplace: an AI agent runs the top-100 real user queries weekly, judges ~50 results per query for relevance, and tracks precision over time — replacing a manual annotation effort nobody could afford.
A daily computer-use agent that sweeps news sites and social platforms for mentions of a marketplace brand and its competitors, classifies tone, and writes a report. It surfaced an emerging wave of user complaints on Reddit before anyone on the team saw it.
56 unstructured clips from a Europe trip, edited into a finished film by a pipeline where Whisper finds the speech, local models score 10,000 frames, and Claude Opus acts as creative director. Five iterations to 'I like it'. Open source.
How I built ConvertStatement — a bank statement PDF parser — from scratch in 14 days. The tech, the tradeoffs, and what I'd do differently.
I'm starting a product factory — shipping SaaS tools one at a time, documenting everything along the way. Here's why, and what to expect.