::Rock Auto Parts CRM (crm.rockautoparts.us) runs the operations of a US auto-parts business I co-own, and I built it end to end. The first version was a Replit prototype with early design passes in Lovable and wireframes generated in Relume, which I then rebuilt for production. Agents take customer calls and capture the requirement in the CRM, which keeps every customer record and order history. Each order is tracked through sourcing, payment, shipment, settlement, chargebacks, and refunds, and the customer gets an automated email and SMS at every status change. Managers see attendance, sales, and per-agent performance in the same place, alongside call scores and live leaderboards. An AI voice agent trains the sales team by calling new reps and roleplaying as a customer across 10 buyer personas. It grades each call on 20 dimensions and returns coaching notes within 30 seconds. A transcript analyser runs the same grading on real customer calls and writes the summary, scorecard, and coaching notes back onto the order. In its first 15 days the training agent rated 400 calls, lifted average call quality from 4 out of 10 to 6, and contributed to a 15% increase in sales.