Generative AI Incubate Accelerator: 6-Week Proof of Concept

Explore the viability of integrating AI into your business operations by launching an MVP tailored to your inaugural AI use case. Employ an agile, iterative methodology to validate your organization's capacity to develop a robust AI solution using Azure Open AI.  

FAIR Generative AI Incubate builds your MVP in three iterations by:

• Selecting an existing foundation model
• Defining and setting up the cloud platform
• Preparing data, adapting, and aligning the model
• Fine-tuning and integration of LLM
• Demonstrating, optimizing, and augmenting the model
• Building an LLM-powered application

Iteration 1: Discovery and Design 

Define requirements for the MVP, develop data pipelines, establish the AI platform, and identify success metrics for your MVP. Key deliverables include: 

• Data sources
• Technology architecture
• Selected LLM
• Fine-tuning approach & requirements
• Prompt templates and definitions

Iteration 2-3: Adapt and Align 

Select a foundational model then adapt, align, and fine-tune it to baseline the model’s performance against a specific task. Key deliverables include: 

• Data pipelines and MLOPs
• MVP that is tuned for a specific task
• Reparameterization of the model
• Prompt tuning

Iteration 4-6: Optimize and Augment 

Continue training the AI solution using prompt engineering and adjust temperature and P and K values to refine the model’s performance. Key deliverables include: 

• Prompts, temperature and model P and K values to produce desired results
• Prompt engineering and training to improve performance against a specific task
• UI that makes it easy to interact with the inference
• Gaps to close before industrializing the AI