License Request for Hybrid-Cloud AI and LLM Platform for Wet Design

Prepared by: Matt Salomon Date: 09/24/2025

Project Phases and Timeline

Phase 1: Concept Phase (Current Stage - Months 1-2)

Current Status: We are currently in the concept phase: Gather datasets, conducting requirements analysis and system design.

What We are Currently Working on:

Gathering existing Wet Design business processes and workflows that may benefit from AI enhancement (For example a risk assessment process can be created using advanced forecasting and gradient boosting methods to address potential part shortages in the future based on provider tendencies and future need forecast)
Historical project data, client communications, and design documentation (csv, excel, doc, docx formats) accumulated over decades of manual operations
Identifying current manual processes that can be automated through AI integration
Gathering information to provide to a chatbot as a RAG process or a fine-tuned chatbot for event more enhancement
Requesting team subject matter expertise in design and engineering workflows that we can feed into our AI models
Identified random other pain points where LLM integration can provide immediate value (such as finding the correct location of a specific nut on the shelves, location of a certain design on the floor design sheets. What could be found in any specific design on our floor. Which example the executives should demonstrate if the client had certain needs.)

Data Assets Available:

Legacy project documentation and specifications
Client requirement databases and communication histories
Design pattern libraries and template repositories
Process documentation and workflow definitions
Quality assurance checklists and compliance requirements
Vendor and supplier interaction logs
Time tracking and resource allocation historical data

AI Integration Layer Design:

The proposed system will create an intelligent mid-layer that:

Connects existing Wet Design systems to cloud-based or on-premise AI services
Processes document ingestion through Azure Document Intelligence
Enables semantic search capabilities across all company knowledge assets
Provides context-aware responses using fine-tuned LLM models
Maintains data security and governance within Azure’s private cloud environment
Integrates with current project management and client communication workflows

Phase 1 Deliverables:

Finalized technical architecture and system design
Data mapping and integration requirements specification
Security and compliance framework documentation
Detailed implementation roadmap and resource requirements

Timeline: 8-10 weeks to complete concept phase and move to prototype

Phase 2: Prototype Phase (Months 3-6)

Objectives: Build and validate the core AI platform functionality with real Wet Design data and workflows.

License and Infrastructure Requirements:

All licenses and services detailed in this document must be procured at the start of this phase:

Microsoft Azure subscription with OpenAI Service access
Google Cloud Platform services for data storage and analytics
Development environment setup and configuration
Security protocols and data governance implementation

Phase 2 Activities:

Data migration and preprocessing pipeline development
Initial model fine-tuning using Wet Design’s proprietary datasets
Core AI integration layer implementation
User interface and API development for internal team access
Security testing and compliance validation
Performance optimization and cost monitoring setup

Expected Prototype Capabilities:

Document processing and intelligent search across company knowledge base
Context-aware query responses for project-related questions
Automated analysis of client requirements and project specifications
Integration with existing Wet Design tools and workflows
Basic reporting and analytics dashboard for usage monitoring

Success Metrics:

96% accuracy in document classification and information retrieval
88% reduction in time spent on routine information searches
Successful integration with at least 3 core business processes
Positive user feedback from internal team testing
System stability and security compliance validation

Timeline: 16-18 weeks to complete prototype and demonstrate business value

Phase 3: Production Release Phase (Months 7-12)

Objectives: Scale the validated prototype to full production deployment across all Wet Design operations.

Production Deployment Requirements:

Migration from development to production cloud or in-house GPU infrastructure (Cloud for LLM, in-house for risk models)
Comprehensive user training and change management program
Full integration with all relevant business systems and workflows
Advanced monitoring, logging, and maintenance procedures
Disaster recovery and business continuity planning

Phase 3 Activities:

Production environment provisioning and configuration
Comprehensive user acceptance testing with full team participation
Advanced model optimization and performance tuning
Complete workflow integration across all departments
Advanced analytics and reporting capabilities deployment
Documentation and training material development

Production Capabilities:

Enterprise-grade AI-powered document management and search
Intelligent project assistance and decision support tools
Automated client communication and proposal generation capabilities
Advanced analytics and business intelligence integration
Mobile and remote access capabilities for distributed team
Full audit trail and compliance reporting features

Success Metrics:

95%+ system uptime and reliability
90% improvement in information retrieval efficiency
Full adoption across all team members and departments
Measurable ROI through time savings and improved client outcomes
Client satisfaction improvements through enhanced service delivery

Timeline: 20-24 weeks to achieve full production deployment and optimization

Total Project Timeline: 12-15 months from concept to full production

Management Visibility Points:

End of Month 2: Concept phase completion and prototype go-ahead decision
End of Month 6: Prototype demonstration and production planning approval
End of Month 12: Full production deployment and ROI assessment

LLM Cloud Infrastructure for ChatGPT Fine-tuning:

Duration: Pay-as-you-go

Considerations:

In my past experience GPT-4.1 mini is the sweet spot for cost and performance efficiency.
We can also use GPT-4.1 nano for semi real-time applications that need low latency.
The costs in this license request sheet are only an estimate. I had not contacted Microsoft and Google sales team yet.
We should present a use-case to the team about why fine-tuning is necessary and vanilla LLM models don’t achieve the requirements.
We can deploy this system reasonably fast within wet design and start the fine-tuning work.
Since the GPT endpoint remains within AzureOpenai instead of API based Openai, we can perform our own data governance and have a highly secure system that wouldnt expose Wet desing data to external entities.
This architecture is future-proof and uses the best technologies as used by major corporations and suggested by data science major players.

Fig1. Proposed System design diagram

Licenses that we need:

Microsoft Azure Subscription Requirements:
- Azure Document Intelligence service license
- Azure OpenAI Service access (with special approval requirement highlighted)
- Azure AI Search service license Clarification that there are no per-user licensing fees
Google Cloud Platform Requirements:
- Google Cloud Storage API access
- Google BigQuery service license No per-user licensing fees

Estimated Monthly Costs (Usage-Based):

High-Performance: $500-2,000/month - GPT-4.1 Full with PTU hosting
Balanced: $50-400/month - GPT-4.1 Mini with pay-as-you-go
Cost-Optimized: $30-200/month - GPT-4.1 Nano with pay-as-you-go

Note: Costs depend heavily on actual usage volumes (pages processed, tokens consumed, search queries)

Recommendation: We can start with Balanced scenario using GPT-4.1 Mini, monitor usage, and adjust.

Service Pricing Models (Usage-Based)

Azure OpenAI GPT-4.1 Models

All pricing per 1 million tokens (regional variations apply):

Model	Input Tokens	Output Tokens	Cached Input (75% off)	Best For
GPT-4.1 Nano	~$0.10/1M	~$0.40/1M	~$0.025/1M	High-volume simple tasks
GPT-4.1 Mini	~$0.40/1M	~$1.60/1M	~$0.10/1M	Production apps
GPT-4.1 Full	~$2.00/1M	~$8.00/1M	~$0.50/1M	Maximum accuracy

Batch API Discount: 50% off standard rates for asynchronous processing

Azure Document Intelligence (OCR)

Pricing: ~$1.50 per 1,000 pages processed

Usage estimate needed: How many pages per month?
Example: 10,000 pages/month = $15/month
Free tier: First 500 pages per month free

Azure AI Search

Pricing: SKU-based per Search Unit (SU) per hour
Basic: ~$0.10/hr per SU × 730 hours = ~$73/month (50GB capacity)
Standard S1: ~$0.33/hr per SU × 730 hours = ~$241/month (160GB capacity)
Cost scales with: Replicas × Partitions × Base SU cost
100GB requirement: Fits in 1× S1 Search Unit

Google Cloud Storage (1TB)

Pricing: ~$0.020-0.026 per GB per month (regional variance)
Estimated cost: $20-26/month for 1TB
Additional costs: Operations charges, egress fees (region-dependent)
Usage estimate needed: Monthly operations volume and data transfer

Google BigQuery (500GB)

Storage: ~$0.02 per GB per month for active logical storage
500GB storage cost: ~$10/month (first 10GB free)
Query processing: $6.25 per TiB of data processed (1TB/month free on many accounts)
Usage estimate needed: Monthly query volume in TiB

Procurement Requirements

Microsoft Azure Subscription

Services: Document Intelligence, OpenAI Service, AI Search
Critical: Submit Azure OpenAI access application
Billing: Pay-as-you-go based on actual usage
Budget range: $50-2,000/month depending on usage and model selection

Google Cloud Platform Account

Services: Cloud Storage, BigQuery
Billing: Pay-as-you-go based on storage and query usage
Budget range: $30-50/month for baseline storage + queries

User Access

4 user accounts with appropriate permissions
No additional per-user licensing costs

Key Decisions & Cost Drivers

Critical Decision: Model Selection & Usage Pattern

GPT-4.1 Model Choice: - Nano: 4× cheaper per token, suitable for simple tasks - Mini: Balanced performance, fine-tuning support - Full: Maximum accuracy, PTU hosting available for guaranteed capacity

Usage Pattern Impact: - High token volumes: Consider PTU hosting for predictable costs - Batch workloads: Use Batch API for 50% savings - Repeated prompts: Leverage 75% caching discount

Cost Variables to Monitor

Token consumption (biggest variable cost)
Document processing volume (pages per month)
Search query patterns (affects AI Search scaling)
BigQuery analysis frequency (TiB processed monthly)

Cost Control

Usage alerts on Azure and Google Cloud accounts
Monthly budget limits per service
Automatic scaling policies for AI Search
Regular usage reviews to optimize model selection

Ordering Process

Azure Services

Contact: Microsoft Account Manager
Special approval: Submit Azure OpenAI Service access request immediately
Setup: Pay-as-you-go billing account
Timeline: 2-3 weeks for OpenAI access approval

Google Cloud Services

Setup: Create account with corporate billing
Enable APIs: Cloud Storage, BigQuery
Timeline: 1-3 business days

Budget Planning Scenarios

Based on common usage patterns:

Light Usage (Pilot/Development)

5M input + 2M output tokens monthly (Mini)
5,000 pages OCR
Basic AI Search
Estimated monthly cost: $80-150

Medium Usage (Production)

20M input + 10M output tokens monthly (Mini)
20,000 pages OCR
Standard AI Search
Estimated monthly cost: $300-500

Heavy Usage (Enterprise Scale)

50M+ tokens monthly
50,000+ pages OCR
Premium AI Search with replicas
Estimated monthly cost: $1,000-3,000+

Appendix: Optional Hive Analytics Platform for Big Data

Additional cost: $2,688/month ($32,256/year) for managed Hive capabilities

Dataproc cluster: $226/month (usage-based)

Metastore Enterprise: $2,462/month (24×7 operation)