Custom Generative AI Platform

The vendor required to provide custom generative ai platform that leverages the existing business relationship with microsoft and our infrastructure investment in the microsoft azure cloud platform to create a generative ai platform with access to private instances of frontier large language models (LLMS) such as those produced by open ai, anthropic, and google.
- The system shall provide faculty, students, and staff with a chat-centric web-based user interface, direct API access for researchers and others developing custom applications, as well as the ability to access LLMS trained and augmented using university-managed high-performance computing infrastructure.
1. General requirements
•    Runs in existing university managed microsoft azure tenant.
•    Follows azure AI foundry best practices.
•    Offeror shall provide agency access to product source code with ability to develop and deploy extensions or modifications.
•    Maintain version history of product with the ability to roll back to previous versions.
•    Allow for system prompt optimization to improve user experience, especially for university-centric queries.
•    Self-service retrieval augmented generation (rag) support. 
•    Allowing faculty and students to provide their own documents and data to supplement the selected LLM.
•    Multiple data source support for rag allowing self-service direct connection to various platforms (e.g., microsoft OneDrive, google drive, amazon s3).
•    Centrally managed access to selected university data sources (e.g., snowflake, SQL databases).
•    Allow users to search and pull data from sources on the web.
•    Image generation (e.g., based on model selection), with capabilities for image editing and in-painting.
•    Video generation (e.g., based on model selection).
2. Monitoring and management requirements
•    Allow for federating and managing LLM resources and related configuration on azure.
•    Ensure resiliency and high availability of the ecosystem for all users.
•    Monitor model usage with near-real time reporting and dashboards.
•    Apply quotas for model usage for rate limiting and cost control.
•    Monitor and audit user queries and user-provided data sources.
•    Archive and purge all history and other data associated with a particular user (e.g., for account deprovisioning).
3. Reporting and billing support requirements
•    Detailed reports of usage across multiple dimensions (e.g., user, group, department, model, time, etc.).
•    Integration to azure cost management.
•    Reports to allow chargebacks to university departments under a microsoft campus agreement.
•    Ability for users to obtain reports of their own usage and associated resource consumption.
4. User interface design requirements
•    Chat-based interaction that will be intuitive and natural for existing users of commercial offerings such as chat GPT.
•    Conformant with wcag 2.1 aa and usable by all members of our community.
•    Customizable theme and branding.
•    Responsive web interface that supports a variety of devices, screen sizes, and input mechanisms.
•    Store chat history on a per user basis. this shall include the following capabilities:
o    User history management (e.g., to remove unwanted history)
o    Set system level controls on the lifetime of chat history stored for users
o    “Incognito” mode for users that does not retain history
•    Export chats or artifacts created into other file formats (e.g., draft into a word doc).
•    Users shall be able to search across multiple previous conversations.
•    Customizable UI theme or persistently set other user interface preferences.
•    Allow users to set custom prompts to instruct how they prefer the generative AI model to respond or to create and organize chats into projects with their own custom instructions.
5. Security requirements
•    Single sign-on through microsoft Entra id.
•    Role-based and group-based access control via grouper.
•    Capabilities for guardrails to prevent abuse, prompt and response filtering.
•    User roles and groups to authorize features such as rag deployment, rag consumption of centrally managed data sources, API access, etc.
•    All user-provided and university data remains completely within the university’s care, custody, and control. 
•    No queries or data supplied to the product and service shall be used to train models for use by parties other than agency.
6. Support requirements
•    Ongoing technical support and maintenance of the product.
•    Escalation path for user support issues with defined response times. 
•    Optional support path for feature requests and enhancements with defined response times.
•    Monthly or quarterly support calls with agency administrators.

- Contract Period/Term: 2 years
- Questions/Inquires Deadline: November 21, 2025