built the keyword pipeline from scratch — dataforseo, drag filters, category browser, 43k keywords for $8

started the day with a curl command and ended it with a full local research tool. the keyword pipeline fetches real google ads keyword data from dataforseo by category code, applies server-side filters before billing starts, and surfaces results sorted by estimated cost per conversion. the core insight driving the whole design: filters above the run bar get sent to the API (you pay less), filters below run in the browser (free). that tradeoff is the entire product.

built in a single index.html plus a server.js CORS proxy. no framework, no build step. ran 368 categories with tight filters — CPC $1-9, competition 0.1-0.5, transactional and commercial intent, volume over 10k — and got 43,667 qualifying keywords for $8.05 total. the biggest single finding from the data: myers briggs personality test keywords at $1.65 CPC across 1.4M monthly searches. that cluster surfaced completely unprompted. it's probably the clearest signal for a next product in the whole dataset.

other things built today: category browser with 3,182 categories classified as digital/physical/both, manually reclassified 2,216 "both" categories with a proper type picker, negative keyword management with phrase vs word matching, saved searches and category selections that survive page refreshes, a monitor list for starring keywords, export CSV (filtered and all), a landing screen that restores previous sessions, and a $/conv metric that replaced the less intuitive conv/$100 framing.

the API key was hardcoded for most of the session and only stripped at the end before the repo was pushed. added to rules: never hardcode keys in source files, even local tools.

repo is live at github.com/artluai/niche-research. next session: deploy as a multi-user web app with auth and subscription billing.