Work · 6 projects
Agents and automation systems built and run against my own operations first. Every entry links to live software or public source — the numbers come from logged executions, not demos.
Reads a customer review, escalates anything serious to the owner and drafts the rest for one-tap approval. It never posts on its own. Verified across three live executions averaging 3.48s — a 5-star queued, a food-poisoning-plus-legal-threat escalated, a child-illness complaint escalated.
Extracts quote details from an inbound email without inventing numbers, and escalates anything touching money or lawyers to a human. 3.19s per run against live executions. The constraint that matters is what it refuses to do unsupervised.
Turns one piece of long-form work into platform-native posts, and gates its own output against brand-voice rules enforced in code rather than in a prompt. If the draft breaks a rule, it does not ship.
A prospecting pipeline that scrapes, scores each prospect 0–100 against an ideal customer profile, personalizes a first line, and stages the outreach. The open-source core replaced roughly $350 a month of SaaS with about 600 lines of dependency-free Python — live OpenStreetMap extraction, SQLite persistence, an audit log. The public demo dashboard runs on simulated data, and the page says so.
Nine-node pipeline: form intake, LLM scoring against a rubric, hot/nurture routing, a personalised reply, an owner alert and a CRM write. 7.4 seconds end to end, verified across four logged executions. Ordering is enforced structurally — the CRM write is a graph dependency of the send steps, so a failed email can never lose the lead.
Chrome side-panel agent that studies a business's site and designs the custom chat agent it actually needs. Bring your own key — runs on Ollama, OpenRouter, or any OpenAI-compatible endpoint, so nothing is locked to a vendor.
More projects being documented. Building in public.