Hi, I'm Blacksmith. I build AI-powered products from zero to one.

Gmail → AI classification → Notion pipeline built in n8n: incoming mail is classified by an LLM and routed into three Notion databases, with de-duplication for safe re-runs and an error branch that logs failures to disk. Watch the 2-minute demo — all inbox data is fictional.

Brand-styled operational spreadsheets that run real businesses: a teacher scheduling & payroll pipeline (7 linked sheets, cascading dropdowns, live dashboard), a cleaning company management system, an equipment breakdown tracker, and a sushi kiosk order manager. Protected formulas, input validation, zero formula errors.

Keyboard-driven desktop tool for triaging large photo dumps: single-pass keep/drop flow, hash-based duplicate detection, and a three-layer delete safety net so nothing is ever lost by accident. Free and open source on GitHub.

Field-service reporting automation: job form data and site photos go in, Claude drafts a structured, branded service report, and a manager approval flow handles approve / reject-with-notes / auto-rewrite / escalation before the final report is emailed out.

Built an AI learning platform from scratch in 20 days — product design, content curation, frontend development (Astro + Tailwind), and deployment. Serving users who want to learn LLMs step by step.

A .pmmap file format + MCP Server that lets AI coding assistants understand product requirements before writing code. Backed by research across 80+ product frameworks and 300+ real PRD documents.

Quantitative stock screening system: cigarette-butt strategy, convertible bond double-low screening, LOF premium arbitrage monitor. Built with Python + akshare + tushare.

Mathematical framework modeling how LLMs reconstruct high-dimensional thinking from low-dimensional text. 16 rounds of experiments completed with statistically significant causal verification (p<0.001).