Docs / 02-requirements/002-onprem-ai-infrastructure
REQ-002 On-premise NVIDIA AI infrastructure for ~50 daily users
Full on-prem AI platform on NVIDIA GPUs — LLM serving, vector DB, RAG, orchestration, UI, security, monitoring, backup, model lifecycle — sized for ~50 daily users.
REQ-002 — On-premise NVIDIA AI infrastructure for ~50 daily users
Need (what is being asked)
A complete on-premise AI infrastructure: local LLM serving on NVIDIA GPU hardware, vector database, knowledge retrieval, orchestration, user interface, security, monitoring, backup, and model lifecycle management, supporting ~50 daily users.
Context (why — what problem it solves)
No AI hardware exists today. Defense/tender business and customer data create a strong confidentiality case: company data must not leave the premises. On-prem also caps recurring cost at predictable levels.
Current state (how it is done today)
No AI capability. Server-room capacity, power, and cooling at the Piraeus HQ to be verified (technical/IT questionnaire).
Success criteria
- Platform in production by ~M6 gate; ≥99% availability during business hours; p95 chat latency acceptable at 10–15 concurrent users; dev/staging/prod separation; tested backup/restore.
Dependencies / related
Blueprint: 05-technical. Related: REQ-003 (bilingual), REQ-008 (security). Board hardware-budget approval gate (M2, roadmap).
History
- 2026-07-03 — captured (source: CEO transformation brief)