1. Mission & System Philosophy
ShunopsAI is a self-healing, multi-provider autonomous operations platform designed to unite edge computing, multi-cloud redundancy, and generative AI reasoning. It eliminates single-point-of-failure vulnerabilities in AI pipelines by orchestrating seamless failover cascades across independent model architectures and hardware accelerators.
2. Technical Infrastructure Stack
- Compute Runtime: Node.js 20 LTS running Fastify v5 on Google Cloud Run (asia-south1) with auto-scaling to zero.
- Database Layer: Neon Serverless PostgreSQL with WebSocket connection pooling and instant branch-first staging.
- In-Memory Cache & Distributed Quota: Redis 7 running on Oracle Cloud Infrastructure (OCI) Compute with AES-authenticated persistent pipes.
- Edge & Bot Defense: Cloudflare Turnstile bot deterrence, client-side Proof-of-Work SHA-256 HMAC challenges, and Cloudflare Workers global distribution.
3. AI Model Hierarchy & Fallback Cascade
All inference paths adhere to the Multi-Provider Fallback Cascade Rule: a failure in any individual provider or model endpoint never terminates an active user session or background audit.
1. Google AI Studio Pool
Primary high-throughput reasoning tier utilizing active non-lite Gemini models:
- gemini-3.8-flash (Primary Active)
- gemini-3-flash-preview
2. Groq LPUs
Ultra-low-latency tensor stream processors specialized for fast tool-calling and AST security audits:
- openai/gpt-oss-120b (120B AST Reasoning)
- qwen/qwen3.8-27b (Fast Tool Calling)
3. OpenRouter Heavyweights
High-capacity fallback chain supporting massive context windows:
- nvidia/nemotron-3-ultra-550b-a55b:free
- meta-llama/llama-3.3-70b-instruct:free
- cohere/north-mini-code:free
4. Multi-Service API Key Pool Anti-Contention Invariant
When multiple distributed services (SaralGati, Kanban Cloud, ShunopsAI) share the centralized master 34-key pool, contention is eliminated via:
- Startup Offset Permutation: Each runner instance generates a randomized startup offset upon initialization, guaranteeing different services begin at distinct indices.
- Zero-Latency O(1) Round-Robin: Pointer advancement strictly executes in O(1) constant time ((index + 1) % keys.length) with zero runtime array shuffling or sorting on hot query paths.
- Instant 429 Pointer Bumping: Upon encountering any HTTP 429 rate limit, the pointer immediately advances to the next key before model fallback.
5. Autonomous Cloud Sentinel & Video Pipeline
In addition to conversational AI, ShunopsAI orchestrates:
- 24/7 Service Sentinel: 12-hour cadence health checks against Cloud Run revisions, Neon databases, and OCI Redis instances, with automatic container reboot triggers upon degraded latency.
- Wan2.1 DiT Video Pipeline: Automated Kaggle Dual-T4 multi-GPU video synthesis for rendering 1080p educational documentaries with edge voice synthesis and auto-publishing to YouTube.