About SignalPipe

Who built this?

My name is Yusuf. I've been building software products for years, mostly backend systems. SignalPipe started as an internal tool I built for myself — a FastAPI backend that monitored Reddit and Hacker News for mentions related to products I was selling. It worked well enough that I decided to package it properly and release it.

The source code for the OpenClaw plugin is public on GitHub (AbYousef739/signalpipe). The backend scoring engine is closed source — the RL weights, LLM swarm configuration, and signal scoring pipeline are what make it work, and that's the part I'm charging for.

Why does this exist?

Cold outreach reply rates have been declining for years. The tools that dominated the market — Apollo, Instantly, Clay — are all built around the same premise: build a big list, send a lot of emails, hope some stick. I was doing this and it was working less and less.

The insight that changed my approach: people who are actively looking for a product post about it publicly. On Reddit, on Hacker News, in forums. They ask for recommendations. They complain about competitors. They describe problems that your product solves. And almost nobody responds to them with genuine help.

SignalPipe automates the detection of those moments. The human still writes the relationship — you review every draft, approve every send. The machine just makes sure you never miss the conversation in the first place.

How does the scoring work?

Every post goes through four stages: a keyword gate that blocks obvious noise, an embedding similarity check against your "anchor sentences" (buyer phrases you define), a sarcasm detection pass, and finally a three-model LLM swarm where a Skeptic, Analyst, and Optimist vote on the signal. The Skeptic has a hard veto — if the most critical judge dismisses it, it never reaches you.

About 85% of posts are filtered before you see anything. What reaches your review queue is the 15% that passed all four stages. Full scoring pipeline explained here.