Rankhog

Build me a Reddit opportunity finder and reply drafter to replace Rankhog, personal scale. Requirements: - A Node CLI plus a local page: node find.js runs the sweep, then a page on 127.0.0.1:3000 lists the hits with a draft beside each thread. - My keywords and target subs live in keywords.json. Sweeps hit Reddit's public JSON endpoints (/r/<sub>/search.json, no OAuth) with a real User-Agent, one request every two seconds. - Keep only the threads already ranking in Google for the keyword, checked via Serper.dev (key in .env), those are the ones the models read back. - Store hits in SQLite (better-sqlite3) with age, upvotes, comment count, and whether my keyword appears in the top comments. Never resurface one I dismissed. - Draft each reply with an LLM (key in .env), fed the thread text and the sub's rules from /r/<sub>/about/rules.json, told to answer first and name my product only where it fits. Store the draft, do not send it. - The page shows title, score, the rule summary, and the draft in an editable box with a copy button and a link out. Everything stays on my machine, localhost only, no accounts, no telemetry. I post by hand, from my own account. - Out of scope: anything that posts, comments, upvotes, or logs into Reddit for me. Automated posting is what gets accounts shadowbanned, say so in the README. - README: the Serper and LLM keys, cost per sweep, and a note to spend a few weeks commenting in my target subs before I mention the product.