I’m building a self-contained AI platform that will handle threat detection and access control for a client’s environment. Everything must live on a Linux/Python stack and run entirely on-premise, so the work spans both software architecture and server planning. Framework choice is still open—I need informed guidance on whether TensorFlow, PyTorch, scikit-learn or another option best matches the workloads we’ll run. Alongside the software, I’ll rely on you for concrete server-setup recommendations: processor/GPU mix, storage layout, redundancy, power, and rack considerations that can comfortably support model training and inference. The timeline points to a start later this calendar year, giving us room to refine requirements and schedule. Being in Central Florida would make white-board sessions easy, though a remote collaborator who keeps close communication works as well. Deliverables • Architectural design covering data flow, model lifecycle, and security hardening • Proof-of-concept codebase deployable on Ubuntu Server, implementing threat detection and role-based access control • Detailed hardware specification sheet with sourcing notes • Deployment scripts (Docker or Ansible) for repeatable installs • Knowledge-transfer session and hand-off documentation Acceptance: the PoC must flag defined test threats with at least 90 % precision and enforce user roles without leak paths. If secure AI systems and hardware sizing are your strengths, let’s talk details soon so the groundwork is ready when the project officially kicks off.