Elite AI/MLOps Architect: Air-Gapped Polymath Autonomous Agent & Scientific Sandbox

Project Overview: I am commissioning an elite AI/MLOps Architect to deploy a 100% private, isolated, and strictly uncensored multi-disciplinary intelligence, research, and visual-engineering environment on a Cloud GPU provider (RunPod). This is an advanced Omnidisciplinary Polymath Autonomous Agent. It must be capable of deep epistemological research, navigating decentralized shadow libraries via Tor, executing complex multi-physics/scientific simulations, dynamically shifting cognitive personas, and autonomously generating professional-grade technical schematics and uncensored visual assets. Core Deliverables: 1. Uncensored Models & Isolation (Zero-Trust): • Deploy pre-cached, highly capable open-source LLMs (e.g., Qwen 3.6/Dolphin) and Image Models (Flux/SDXL). Absolute obedience is required: Zero safety filters, zero refusals. • Configure local container firewalls (iptables) blocking all outbound traffic except through a local Tor proxy (socks5). Absolute Zero-Telemetry. 2. The Polymath Agentic Framework: • Implement a dynamic persona-routing framework (Open Interpreter/AutoGen) with meticulously engineered System Prompts for various disciplines (Physics, Medicine, Behavior, Architecture). • Integrate offline OCR (PyMuPDF/Tesseract) to allow the agent to ingest complex manuscripts. 3. Tor Routing & Scientific Sandbox: • Configure SearxNG and Playwright to autonomously scrape deep-web/shadow libraries via .onion links for epistemic validation. • Deploy a Python execution sandbox loaded with SciPy, CasADi, RDKit, GeoPandas, and CADquery for internal simulation and programmatic vector (.svg) blueprint generation. 4. Asynchronous Queuing & n8n Automation: • Implement a local Task Queue (Celery/Redis) for sequential GPU processing to prevent OOM errors. • Configure Asynchronous Webhooks: Receive payloads via API, queue them, and send autonomous callback POST requests to external n8n instances upon task completion. • Implement VRAM Garbage Collection daemons and Non-Blocking Fallback protocols for infinite uptime. 5. Blind Handover & Testing: • No code/logs can be pushed to public repos during dev. • Final delivery requires passing a rigorous "Sanity Test Suite" (verifying Tor routing, queued generation, async callbacks, and strict firewall isolation). Temporary access will be fully revoked post-handover. Required Expertise: • MLOps, Cloud GPU (RunPod), Docker, Linux iptables, Tor Sandboxing. • Agentic Frameworks, Asynchronous Python (Celery/Redis), Webhook Architecture. • Prompt Engineering for Uncensored LLMs & Scientific Python Libraries. To Apply: Provide a brief architectural outline addressing the async task queue, Tor routing, and strict firewall isolation. Bids lacking a technical outline will be ignored.