Business Problem

• Manual content generation was time-intensive, limiting the platform's ability to scale and personalize content.
• Curating high-quality, contextually relevant images for articles posed significant challenges.
• Delivering tailored travel itineraries and recommendations required substantial manual intervention, limiting user engagement potential.

Solution

GoML collaborated with Curly Tales to deliver a Proof of Concept (PoC) using outstanding Generative AI technologies. The solution addressed these challenges through the following components, supported by an advanced tech stack:

Article Outline Generation
- Technology Used: Claude V3.5 (LLM Model) with Open Search Vector Database
- Developed a system with a rag pipeline to generate structured and detailed article outlines based on user input.
- Added source references

AI-Powered Itinerary Planner
- Technology Used: Claude V3.5 (LLM Model) with Open Search Vector Database
- Built an API endpoint to generate personalized travel itineraries using Curly Tales' extensive content database.
- Provided tailored recommendations to improve user satisfaction and engagement.

Infrastructure
- Technology Used: AWS
- Ensured scalability, reliability, and seamless integration with Curly Tales’ existing systems.

Image Generation
- Technology Used: Stable Diffusion V3 (Image Model)
- Leveraged AI to create high-quality, contextually aligned images that enhance the visual appeal and engagement of articles.

Simple User Interface (UI)
- Technology Used: React
- Designed a lightweight and intuitive UI, enabling users to input topics and keywords and receive AI-generated article outlines and visuals effortlessly.

Architecture

• Data Sources and Knowledge Base Creation
  - AWS S3 Buckets: Stores raw data for processing.
  - Data Mart: Aggregates and organizes data for further use.
  - Knowledge Base Creation: Curates the data to form a centralized repository for vectorization.
• Data Preprocessing
  - Python Native Scripts: Executes preprocessing logic on raw data.
  - Data Preprocessing Layer: Cleans, structures, and prepares data for vectorization.
• Vectorization Process
  - Business Intelligence Heuristics: Applies analytical rules for data transformation.
  - Chucking and Rejoining Algorithms: Optimizes data chunks for embedding.
  - Metadata Processing: Annotates data with relevant metadata.
  - Embedding Model: Converts processed data into vector representations.
  - Storage: Uses OpenSearch Vector DB for storing vectorized data.
• Query Processing
  - User Input: Accepts conversational queries from users.
  - Query Processing Module: Analyzes and processes user queries to extract intent.
• Advanced Retrieval-Augmented Generation (RAG) Pipeline
  - AWS Bedrock: Powers foundational AI services and manages to embed them.
  - Data Pipelines: Facilitates smooth data flow and transformation.
  - Parameter Extraction and Generation Layer: Extracts parameters from queries for relevant response generation.
  - Historical Chat Storage: Stores conversation history for contextual replies.
• Focused AI Agents
  - Multiple AI Agents: Specialized agents handle distinct functionalities based on query parameters.
• Deployment and Infrastructure
  - Code Repository: Centralized version control for all components.
  - Docker: Containerizes applications for consistent deployment across environments.