I need an AI-driven simulator that lets my customer-support trainees practise realistic, German-language conversations without the cost of a commercial platform. The system must handle three everyday scenarios—sales conversations, service-issue resolution, and product-information queries—and respond naturally as the “customer”. After each session the agent should receive a performance score, a concise yet detailed report, and, when desired, live in-call hints so they can adjust their tone or approach on the spot. I want to seed the simulator with true-to-life transcripts that we already own, then let it refine itself over time through a mixed learning approach: automated self-training on anonymised interaction logs, plus occasional manual tweaks from our trainers to keep the language and brand tone sharp. All content, analytics, and UI labels must be in German. Because there is virtually no budget, please build on open-source frameworks or freely available models and hosting. A lightweight web interface or local desktop app is fine as long as it runs on standard company hardware and stores data securely. Deliverables • A working prototype that I can install and demo internally • Simple admin panel to upload new scenarios and export interaction logs • Real-time feedback overlay and post-call analytics dashboard (scores & reports) • Clear documentation covering setup, model fine-tuning, and how trainers can perform manual updates If you have experience marrying open-source LLMs with conversational front-ends—and can keep costs close to zero—let’s talk.