- SignalDesk9 hr ago
Original Summary
Hi HN, I'm Salman. I started ReacherX as a side project in mid 2023 while studying at university, working part-time, and freelancing. I went full-time on it in June 2026. First time sharing something I built on Hacker News ;)<p>ReacherX is an open-source platform for finding the people you're looking for on X/Twitter and LinkedIn.<p>There's also an interactive demo linked in the README, prepopulated with fictional data, so you can click around and get a feel for the product without setting anything up. It isn't connected to the real backend.<p>Ideal use cases include finding potential customers, doing market research/validation before building something, and finding ideal candidates when hiring.<p>I built it because even in 2026, finding the right people/network isn't easy. One question always boggled me: why can't I just describe the people I need in plain English and see a list of people along with the actual sources/evidence showing why they are the right match?<p>Existing tools are focused heavily on sales professionals. I am not a salesperson. I don't understand sales jargon. I wanted something that "just works" when I describe who I am looking for. No campaigns or sequences to learn.<p>After spending three years in this space, I also came to realise that this is less of a funnel problem and more of a networking problem. The whole space is also pretty closed, which is one reason I wanted ReacherX open source.<p>The workflow is roughly:<p>- Describe who you're looking for and why. - The Agent creates ideal profiles and example mock profiles for you to review. - It continuously searches X/Twitter and LinkedIn using generated search queries. - Each person is qualified with an LLM and enriched with supporting sources/activity and other details. - ReacherX only surfaces people with a match score of 70% or higher. - At 90%+, it can generate an outreach plan for review. - DMs, replies, liking/commenting, profiles, voice notes, and media attachments are built in.<p>You can ask the Agent to handle outreach, or do it yourself through the UI. You can also upload a product demo and tell the Agent to use it when appropriate.<p>It has memory and adapts based on search performance and your feedback/preferences stored in memory.<p>The searches run in the backend, so the browser can be closed. First results usually take around 5–30 minutes, and the workflows keep running until plan limits are reached or a backend error stops them.<p>ReacherX doesn't rely on an internal static prospect database for discovery. It performs searches against current X/Twitter and LinkedIn data, so results can be very recent.<p>I ran it for 2–3 weeks while looking for potential customers. It processed 50,000 people, but only around 300 scored 70% or higher and were surfaced to me.<p>It's an end-to-end TypeScript project using Next.js and Convex. ReacherX uses Convex Agent, Workflows, Workpool, Memory/RAG, etc. Those durable workflows are the backbone of the system. ReacherX is also part of the Convex Open Source program.<p>One thing I'm still thinking hard about is how this architecture will hold up at scale. At a high level it's a continuous loop: generate queries -> search platforms -> qualify/disqualify results with an LLM -> enrich matches -> keep searching.<p>Another challenge has been social-data APIs. Before X introduced Pay Per Use, its API was prohibitively expensive for this kind of product, and even now it can still be costly. Third-party APIs made it possible for me to build this at a much more reasonable cost.<p>ReacherX is currently in beta (v4). It has bugs and rough edges, and I'm working on it every day. The hosted product is paid and doesn't have a free tier yet. If you don't want to use the hosted version, you can self-host it. I'm also open to contributors and bug reports on GitHub.<p>I'd especially love feedback on the idea, match quality, architecture/scaling approach, and product philosophy. Anything is welcome :)
- 情报分类：工作与职业机会
- 分类依据：内容涉及招聘、求职或职业发展
- 信息来源：Hacker News 新项目
- 发布时间：2026/9/19 04:55:14
- No replies yet