On-Premises AI Knowledge Management System
A Techno-Commercial Proposal for the Water Solution Company (WSC) to unlock strategic insights from proprietary data.
Executive Summary
Transforming data into a strategic asset for WSC.
This proposal outlines the deployment of a state-of-the-art, On-Premises AI Knowledge Management System. The solution, a Sovereign Retrieval-Augmented Generation (RAG) System, operates entirely within WSC's premises, ensuring absolute data security and sovereignty.
By transforming WSC's vast document library into an interactive knowledge base, this system will empower personnel with instant, accurate, and verifiable answers, fostering faster, more informed decision-making.
Total Investment (Excl. VAT)
SAR 400,859.80
The Proposed Solution
A Sovereign RAG System: Secure, Powerful, and Foundational.
The Challenge of Untapped Knowledge
WSC possesses a vast collection of strategic documents. The manual effort to search this data is prohibitive, limiting timely insights and devaluing these critical assets. Cloud-based AI is not an option due to the sensitive, national-security-related nature of the data.
A Foundation for Future AI
This project establishes a reusable AI Center of Excellence within WSC. The hardware and software infrastructure can be leveraged for future high-impact applications, ensuring a compounding return on investment.
How It Works
1. Ingest & Index
Securely processes WSC documents into a specialized, searchable knowledge base.
2. Retrieve
When a user asks a question, the system intelligently finds the most relevant information.
3. Generate
A locally-hosted LLM generates a concise, accurate answer with citations.
Project Scope & Details
The technical foundation of the AI solution.
System Architecture & Access
The solution is built on a high-performance, single-server architecture designed for intensive AI workloads, equipped with a next-generation GPU and containerized using Docker for stability. For authorized employees, remote access is facilitated exclusively through WSC's existing corporate VPN, the management of which remains the responsibility of the WSC IT department.
Proposed Hardware Supply
| Component | Model Description |
|---|---|
| Case | SilverStone Seta H2 |
| Motherboard | ASUS Pro WS WRX90E-SAGE SE |
| CPU | AMD Threadripper PRO 7965WX |
| CPU Cooler | Noctua NH-D9 TR5-SP6 4U |
| RAM | G.SKILL G5 Neo 128 GB DDR5-6400 CL32 RDIMM |
| GPU | NVIDIA RTX PRO 6000 Blackwell Edition (96 GB) |
| PSU | CORSAIR AX1600i 1600W Titanium |
| SSD (System) | Crucial T705 Gen5 - 2TB NVMe |
| SSD (Storage) | Crucial T705 Gen5 - 4TB NVMe |
Proposed AI & Software Stack
| Component | Recommendation | Justification |
|---|---|---|
| Generator LLM | Llama 3.1 70B | Apex open-source model for reasoning and instruction-following. |
| Embedding Model | Qwen3-embedding-8b | A powerful, large model to ensure accurate semantic search. |
| Re-ranking Model | Qwen3 4B/8B | Critically refines search results to improve precision. |
| Vector Database | ChromaDB / Qdrant | High-performance, specialized database for vector storage. |
| Model Serving | Ollama | Simplifies local management and serving of LLMs. |
| Backend & OS | FastAPI, Docker | Modern, high-performance, and industry-standard tools. |
Scope of Services
- RAG Pipeline & Model Integration
- Custom Backend API & Ingestion Agent Development
- Vector DB Setup & Configuration
- Server Assembly, OS & Driver Installation
- Model Serving & Containerization
- Project Management & Client Coordination
- System Testing, Documentation & User Training
Commercial Proposal
Investment breakdown and payment milestones.
Pricing Summary
Payment Milestones
- 130% Advance Payment: Upon contract signing.
- 230% Hardware Procurement: Upon order confirmation.
- 320% Backend Readiness: Upon functional demo.
- 410% On-Site Deployment: Upon installation & training.
- 510% Final Acceptance: Upon project go-live.
Project Execution
Responsibilities and exclusions for a successful project.
Customer Responsibilities
- Provide secure access to the document corpus.
- Designate a primary technical point of contact.
- Prepare a suitable site for the server (power, cooling, network).
- Ensure personnel are available for consultation and training.
- Manage all user accounts and access control.
General Exclusions
- Ongoing data curation or management post-ingestion.
- Model re-training or fine-tuning beyond initial scope.
- Maintenance of Client's internal IT network or VPN.
- Any hardware or software not explicitly listed.
- Any civil works for site preparation.
Service Level Agreement (SLA)
Our commitment to service and support for one year.
System Availability
Target Uptime
98%
During KSA Business Hours
Sun-Thu, 8am-5pm
Scheduled maintenance will be performed outside business hours with 48 hours prior notification.