cai-exos-systems/daveadmin-exos-demo:exosneeds/README.md
EXOS Agentic BSS Needs
This folder is the EXOS-facing handoff for duplicating and extending the demo at
`demo.exos.bluenotelogic.com`.
The recommended starting point is deliberately modest:
- one VPS for the web app, databases, tools, traces, and vector store
- one hourly GPU rental, starting with a SimplePod-style RTX 4090 instance
- plain Qwen 7B as the baseline LLM
- no tuned LLM required for the first EXOS build
- optional evidence retrieval added as the thesis-system pattern, branded for EXOS as a
controlled evidence layer rather than CaveauAI
Assumption: EXOS has telecom/BSS skills, TM Forum knowledge, delivery discipline, and
support capability, but they should be treated as starting from zero on dedicated
AI infrastructure for this project. The plan therefore includes the VPS, GPU rental,
tooling, observability, retrieval, and Azure migration path from the ground up.
Files - `index.php` is the browser-readable executive and technical summary. - `architecture.md` explains how the current demo produces results. - `infrastructure.md` lists what to install on the VPS, GPU pod, and later Azure. - `pricing.md` gives current starting prices and source links for GPU and VPS options. - `questions.md` lists the decisions EXOS should answer before implementation.
Core Recommendation Start with: - VPS: Hetzner CPX31 or CX42/CX43-class Linux server for demos and development. - GPU: SimplePod-style RTX 4090 hourly rental for Qwen 7B/14B development. - Competitor GPUs: RunPod for reliability, Vast.ai for cheapest marketplace experiments. - Competitor VPSs: DigitalOcean for predictable droplets, Vultr for similar global cloud VMs. - Azure target: move the same architecture to Azure App Service or AKS, Azure Database for PostgreSQL/MySQL, Azure AI Search or Qdrant on AKS, Azure Monitor/Application Insights, Key Vault, and eventually Azure GPU VM or Azure ML managed online endpoint.
Naming For EXOS-facing demos, do not call this CaveauAI. Use: - EXOS controlled evidence layer - EXOS standards evidence layer - EXOS agent validation layer - thesis evidence system The point is to show what the thesis system adds to a plain Qwen 7B agent: evidence, source locking, tool governance, scoring, and auditability. ```
Files - `index.php` is the browser-readable executive and technical summary. - `architecture.md` explains how the current demo produces results. - `infrastructure.md` lists what to install on the VPS, GPU pod, and later Azure. - `pricing.md` gives current starting prices and source links for GPU and VPS options. - `questions.md` lists the decisions EXOS should answer before implementation.
Core Recommendation Start with: - VPS: Hetzner CPX31 or CX42/CX43-class Linux server for demos and development. - GPU: SimplePod-style RTX 4090 hourly rental for Qwen 7B/14B development. - Competitor GPUs: RunPod for reliability, Vast.ai for cheapest marketplace experiments. - Competitor VPSs: DigitalOcean for predictable droplets, Vultr for similar global cloud VMs. - Azure target: move the same architecture to Azure App Service or AKS, Azure Database for PostgreSQL/MySQL, Azure AI Search or Qdrant on AKS, Azure Monitor/Application Insights, Key Vault, and eventually Azure GPU VM or Azure ML managed online endpoint.
Naming For EXOS-facing demos, do not call this CaveauAI. Use: - EXOS controlled evidence layer - EXOS standards evidence layer - EXOS agent validation layer - thesis evidence system The point is to show what the thesis system adds to a plain Qwen 7B agent: evidence, source locking, tool governance, scoring, and auditability. ```