Assessment Phase
- Technical workshops for Information gathering such as current environment, for establishing the Azure OpenAI groundwork, user stories, use cases and flows
- Capacity planning on data sizes, file type and file count on Sharepoint for indexing
- Define environments for Azure OpenAI POC development
- Flow diagram and design mockups
Deliverable: High-Level Architecture Design Document
.
Remediate Phase
- Low-level Architecture Design
- Deploy Azure Infrastructure to support the design:
• Azure OpenAI Services
• Azure App Services
• Azure Storage Account
• Azure Key Vault
• Azure VNET/subnets, VNET peerings
• Azure private endpoints, NSGs, route tables
Deliverable: Low-Level Architecture Design Document
.
Enablement Phase
- Development of chatbot frontend (support both desktop web and mobile web)
• Create a React component that implements the chatbot user interface
• Authenticate the chatbot users with Azure AD and acquire access tokens for calling the search service and other protected APIs
• Integrate the chatbot with Azure AD and enforce role-based access control (RBAC) based on the user’s group membership or claims
• Refresh the access tokens when they expire and handle token renewal errors
• Make HTTP requests to the search service and other APIs using the access tokens in the Authorization header
• Benchmark the performance of the chatbot and search service, and compare it with the expected query load and latency
• Use Azure Monitor to collect and analyze resource logs and metrics from the chatbot and search service
• Identify any issues or bottlenecks that may affect the accuracy or reliability of the chatbot solution

- Development of chatbot backend
"Token limitation
Monitor of chat conversations
Export chat of 20 conversations
Choose chatgpt model
Config chatgpt options – top k & top p, temperature
Conversation log database "
Deliverable: Application ready for test
.
Onboarding Phase
"Implementation will be rolled out in two phases:
* Phase I: For Staff
* Phase II: Mass rollout for students"
- Define UAT acceptance criteria
- Perform UAT testing and provide report
- Up to 4-hour training session to clients over Teams
Documentation handover
Deliverable: UAT Test Report
Deliverable: Azure OpenAI Configuration Summary
Deliverable: End User Guide
Deliverable: Azure Admin & Operations Guide

Assumptions
*All hardware/software provided by Customer with no cost to Hazedawn.
*Integration with 3rd party systems and other on-premises systems are excluded.
*UAT will not cover more than 20 users in total, further user enablement will require project next phase.
*User onboarding include End User Guide and 4-hour training session only.
*Customer already subscribed to Sharepoint services.
*Integration based on Sharepoint Online without on-premises Sharepoint servers.
*For first release will be expected accuracy 60% to 70% on indexed data, to be further optimized and fine-tune in project next phase.
*Out of Project Scope for POC: Accuracy assurance, native mobile app, model fine-tuning, HA & DR designs, security scanning, pen test, load test, IaC, DevOps CI/CD, monitoring and reporting, indexing documents outside Sharepoint, customized interface for search client, advanced search features (faceting, filtering, scoring, synonyms)
*Client will recognize Hazedawn as the PAL (Partner Admin Link). 
*Azure consumption costs are excluded from this SOW and will be borne by the client. 
*Azure Active Directory will be utilized for user management and authentication purposes. 
*A maximum of two user roles are provided: Admin and User. 
*The university's logo can be integrated into the portal. The logo will be supplied by the client.