I've been building a real-time ai copilot for sales reps. every tool i looked at only helps after the call is over

I'm a solo founder who's been building developer tools in public for the past year (open-source linters, cost trackers, etc.). A few months ago I started working on a completely different problem: helping SDRs during live sales calls.

The gap I kept seeing:

I spent weeks talking to SDRs and sales managers, reading hundreds of reviews of sales tools. The entire AI sales tool category in 2026 does the same thing: record the call, transcribe, analyze, report. Gong. Chorus. Avoma. Clari. All post-call.

Post-call analysis is genuinely useful. But it solves yesterday's problem. Your SDR already froze on the pricing question. Already fumbled the objection. Already lost the deal.

The few tools that attempt real-time (Revenue.io, Salesken) are enterprise-priced and built for teams with Salesforce. If you're a 5-50 person sales team without a $50K/year tech stack? Nothing.

SDRs spend only 28-39% of their workday actually selling. During those precious live call minutes, they're completely on their own.

What I'm building:

CueRep. A real-time AI copilot that sits as an invisible overlay on top of Zoom, WhatsApp, Google Meet. It listens to the conversation, matches keywords against your product knowledge base, pulls CRM context, and surfaces the right answer in 1-3 seconds. Before the awkward silence.

The key technical decisions:

• Tauri (Rust + React) instead of Electron: Real-time audio processing at sub-second latency. Electron's overhead made it impossible to hit the response times I need.
• Dual-stream audio: Captures rep mic and prospect system audio as separate streams. No ML diarization needed, which eliminates a whole class of latency and accuracy problems.
• Aho-Corasick keyword engine: Matches prospect speech against battle cards in under 1 millisecond. When a prospect says "what about competitor X," the rep sees the battle card before they finish the sentence.
• RAG pipeline: Semantic search over product KB + prospect context, feeds into Claude Haiku for fast inference. Full answer in 1-3 seconds.
• Pre-call CRM assembly: Pulls prospect context from Zoho/Google Sheets before the call starts. The rep doesn't open the call cold.

Where I am:

Still building. The waitlist site just went live at cuerep.com. No launch date. I'm being honest about that. I'd rather ship it right than ship it fast.

This is the first time I'm talking about it publicly. I've been building in public with my dev tools, but CueRep needed more quiet building time because the audio pipeline is genuinely hard.

If you're an SDR, BDR, or manage a sales team: what kills you the most during live calls? Not the prep before or the analysis after. The actual live call. What breaks down?