Projects that solve real problems.

Challenge: Competitive intel tools are bloated SaaS that hoard your data and never quite answer the question you asked.

Solution: Local-first pipeline: ingest → LLM classify → summarize → brief. Durable storage on Turso, cron on Railway, briefs committed to git.

Win: Briefs I actually trust, fully owned data, model-swappable via OpenRouter. Open-sourcing once the README is honest.

Challenge: Every Claude Code session started the same way, re-explaining the same decisions, re-correcting the same mistakes, re-pasting last week's context.

Solution: A persistent memory layer Claude reads at session start. Corrections, decisions, and patterns saved to disk so context survives past the conversation window.

Win: Shipped to npm (npm i -g apsolut-cortex), drives my daily AI workflow, and stops Claude from forgetting what we agreed on last Tuesday.

Challenge: Buyers ask LLMs for category recommendations and brands either get named or they don't. No incumbent tool actually answers frequency × position × sentiment × citation source, across providers, over time, at a sane price.

Solution: Multi-provider scan engine (one OpenRouter key, all platforms) → temporal mention extraction → sentiment & competitor radar → content-gap brief generator. 24-table schema, auth, scan dispatch, OpenRouter adapter all wired.

Win: Marketing teams see exactly where they stand in AI answers, frequency, position, sentiment, sources, competitor delta, and get the content brief that closes each gap. Without enterprise pricing.

Challenge: Useful prompts scatter across Slack DMs, Notion pages, and old browser tabs. By the time you find the right one, you've already retyped a worse version.

Solution: Personal prompt vault, categorized snippets with multi-field copy, search, one click to clipboard. Lightweight, fast, runs in the browser. Next.js + Turso.

Win: Three seconds to the right prompt instead of three minutes hunting through last quarter's Slack archive. Open it every day.