work / Chrome extension · lead intelligence
Lead Surf
A LinkedIn lead-intelligence Chrome extension that reads your feed, classifies buying signals in real time, and surfaces the leads worth acting on, without automating replies.
01 / the problem
Why it exists
SDRs, agency owners, and consultants miss buying signals buried in their LinkedIn feed: someone complaining about a competitor, a hiring signal, a client posting. Reading 150-200 posts a day by hand doesn't happen.
02 / what it does
The product
- ▸The MV3 Chrome extension scans the feed on a schedule (roughly every 2 hours), pulls about 150-200 posts, and classifies each into intent categories: lead / competitor mention / hiring signal / client activity / neutral / chatter.
- ▸It produces a daily brief, a per-lead dossier, an outreach tracker, and reverse-engagement views. It never auto-replies, a deliberate product principle: be the intelligence layer, not a spam bot.
demo
See it run
03 / architecture
How it works
01 · Structured JSON classifier
GPT-4.1-mini returns a strict JSON object with a compact schema. Categories are constrained, and client_activity is assigned in code, not by the model.
02 · Deterministic watchlist override
The Clients badge is only applied when the post's author actually matches the user's watched-client list, a lookup, not an LLM guess. Same principle as Ganit's price math.
03 · Voyager API from the browser only
The extension calls LinkedIn's internal Voyager API directly from the client, so LinkedIn cookies never touch the server. Calls are paced and deduped so the activity looks like normal browsing, not a scraping burst.
04 · BrightData for bulk enrichment
Location enrichment runs through BrightData instead of Voyager, specifically to avoid registering about 150 profile views per scan, which would flag the account.
05 · Prompt-injection hardening
Third-party post text is untrusted input flowing into an LLM, so generators that touch user or third-party text are hardened against injection.
04 / principle
The intelligence layer, not a spam bot
Lead Surf never auto-replies. The value is knowing who is worth talking to and why; the outreach is the user's call. And like Ganit: don't ask the model to judge what you can look up.
05 / stack & outcomes
Stack
Outcome
Shipped to the Chrome Web Store with a marketing site. Evolved from Keyword Scouter: beta users found multi-platform too scattered, and that feedback is what focused it on LinkedIn. I killed my own feature based on user feedback.