Jay Gould

SiftPost is the latest project I’ve been building — a human-in-the-loop social monitoring tool that helps founders find Reddit, Hacker News, and X posts where their product is genuinely relevant, and engage with them in a way that feels human rather than spammy.

The idea came from watching how most “Reddit marketing” tools encourage stealthy self-promotion — fake accounts, undisclosed plugs, drive-by replies. That approach burns accounts, annoys communities, and rarely converts. I wanted to build something that takes the opposite stance: help founders become real participants in the communities where their customers already hang out, and only promote their product at moments where it’s actually useful.

How it works

When a user signs up, they describe their product in plain English. SiftPost uses Claude to suggest keywords for each platform and to generate a “persona” — a set of platform-aware tone guidelines, since the way you write on Hacker News is very different from how you write on Reddit.

From there, two things happen in parallel:

• Backfill — SiftPost pulls the last 7 days of relevant Reddit posts and the last month of Hacker News posts via the public search APIs, generates embeddings, and scores each post for relevance, intent, and whether the thread is still worth engaging with. This gives users immediate value on day one rather than waiting around for new posts to appear.
• Live monitoring — BullMQ workers poll Reddit, the Algolia HN Search API, and a Twitter search API every few minutes for new matches. High-scoring posts get a draft reply written by Claude and surfaced to the user.

Every opportunity is classified as either an Engage (a relevant thread where a genuinely helpful, non-promotional reply makes sense) or a Promote (a strong fit where the product solves the author’s problem, with honest disclosure built into the draft). This dual strategy is the core of the “ethical marketing” angle — it nudges users toward building account credibility instead of burning it.

Tech

The backend is a Node.js + TypeScript Express API with Postgres and pgvector for storing posts and running semantic similarity searches. BullMQ on Redis handles the polling, scoring, and notification queues. Claude does the scoring, drafting, and persona coaching, and OpenAI’s text-embedding-3-small model generates the vector embeddings (Anthropic doesn’t currently offer an embeddings model, so this is the one place OpenAI sneaks in).

The dashboard is a Next.js client styled with Tailwind, and the marketing site is a separate Next.js landing page. Notifications go out via Twilio SMS to start, with WhatsApp planned as a premium channel.

It’s still early, but the core pipeline — fetch, embed, score, draft, notify — is live across all three platforms. If you’re a founder trying to find your first users without resorting to spam, give it a try.