Architectural View on Enterprise Level Implementation for Gen AI solutions

By Kamal Atreja - Head of Delivery - Ubique Digital

As per the Gartner research “By 2026, more than 80% of independent software vendors?will have embedded GenAI capabilities in their enterprise applications, up from less than 5% today.?As the generative AI market disrupts both technology builders and users, it simultaneously fuels a wealth of opportunities to design and deliver impactful GenAI features.”

Architecture has become a focal point in implementing Generative AI solutions across a wide range of use cases. At the enterprise level, where security and governance are paramount, it is crucial to establish a robust, generic solution or design pattern that addresses diverse scenarios within a unified framework. This article explores architectural considerations for integrating Generative AI solutions and building them with confidence.

Pre Read:

To gain a solid understanding of the concepts discussed here, we recommend reviewing the following foundational articles:

• All About Generative AI, LLM and Transformers
• Generative AI Deep Dive
• RAG and RAG implementation

By mastering these fundamentals, you will be better equipped to explore the architectural landscape of Generative AI.

Generative AI Solution Overview

Before diving into the architecture of a Generative AI setup at the enterprise level, it’s important to first understand the key modules and aspects that form the foundation of a Generative AI solution. These modules, which serve as the building blocks for applications and usability, can be broadly categorized into three main groups:

1. Content Consumption - This includes applications, users, features, and use cases where the generated content is utilized.
2. Content Generators - The core systems and models responsible for creating content based on the defined input and requirements.
3. Technology Enablers - The underlying tools, infrastructure, and technologies that facilitate and support the entire Generative AI ecosystem.

When designing a solution, the major entities can be organized under the three major categories outlined earlier:

A. Content Consumption

These are applications, users, features, and use cases that utilize the generated content:

• Websites, Applications
• Requesting Users
• Requesting Modules
• Prompt Generators/Testers

B. Content Generators

Core components responsible for generating content:

• Large Language Models (LLMs)
• Artificial Intelligence Modules
• Foundation Models
• Model Gardens

C. Technology Enablers

The infrastructure and tools that support Generative AI systems:

• Vector Databases/Embeddings
• Application Frameworks or LLM Frameworks
• Retrieval-Augmented Generation (RAG) and RAG Integrations
• Data Science Validation/ML Setup Validation
• Data Hub/Data Lake/Data Availability
• Security
• Governance
• Logging and Monitoring
• MLOps/DevOps

Solution Design and Architecture: Generative AI in the Travel Industry

Let’s explore a solution design for the travel industry by creating a targeted upsell summary and interaction module. This includes input capabilities for activities, new adventures, excursions, cruises, or specific consumer needs in a B2C context. The output serves two main purposes:

1. Customer-Facing: Display recommendations on a travel website to engage users.
2. Agent-Facing: Assist customer care agents in upselling plans to customers based on their history.

The solution is designed as a simple add-on to an existing website or customer care portal. It aims to:

• Attract Customers: Provide personalized suggestions based on preferences.
• Increase Revenue: Upsell activities tailored for families, adventure seekers, or fun-loving individuals.

This approach can also be applied to retail websites to target specific audiences and increase upsell opportunities.

Content Originators and Data Flow

• Content Originators: Travel websites and a customer care agent web portal act as primary interfaces.
• Data Warehouse: All travel plans, customer data, and purchase history are stored in the data warehouse. This includes information on:

Travel history and preferences (e.g., outing types)

Buying patterns, hot-selling activities, and deals.

Key Processes

• Data Enrichment:

Customer history, preferences, and buying patterns are combined with relevant travel plans from the data warehouse.

• Prompt Augmentation:

Contextual prompts are generated for the LLM (Large Language Model) using RAG (Retrieval-Augmented Generation).

• Web Scraping:

Updates on the latest plans from the travel website are scraped and stored in the RAG Vector Database to ensure information accuracy and freshness.

• Generative AI Integration:

The augmented prompts, enriched with contextual data, are fed into the LLM.

The LLM generates responses tailored for upselling.

• Response Delivery:

The LLM outputs are returned to the travel website or the customer care portal with contextual and personalized views, enhancing the overall user experience.

This architecture provides seamless integration, enabling both customers and agents to leverage Generative AI for tailored travel recommendations and upselling opportunities.

To learn more about how Gen AI can be used in your organisation, contact us Ubique Digital LTD