A sales-intelligence agent platform

Cartley is a sales-intelligence platform: agents that research prospects and run playbooks so a sales team starts each conversation already informed. This was our first agent-platform build, not a brochure site, and the bar was to ship something that actually runs reliably, not a demo.

The hard parts were the ones users never see: keeping agent runs idempotent, holding a run-cap window correctly across services, and ingesting lead data cleanly enough to act on.

A multi-agent platform that runs research playbooks against prospects, with a lead-discovery ingest pipeline feeding it. We built it on Neon and Vercel, ran a pilot to harden the run-cap and clock-skew logic, and architected the streaming layer to decouple from agent compute so it scales past a single user.

It's a real system doing real work, not a mockup.

Playbooks are defined as data, not code, so new research strategies ship without a deploy. A team view tracks per-rep activity, and every run lands in an auditable history.

Sign-in runs on Clerk, so sessions, multi-factor, and account security are handled by a dedicated auth provider instead of hand-rolled. The unglamorous part, done right from day one.

The platform ran a live pilot and is in active development. This is an engineering case study: it shows how we build AI systems that actually run, not a finished client engagement.