Business Problem

GroupSolver faced key challenges in improving its data handling capabilities:

• Needed AI-enhanced data extraction and summarization to generate efficient insights.
• Sought to improve data summarization quality to provide more actionable insights.
• Required a streamlined, efficient information processing solution to enhance research workflows.

Solution

To help GroupSolver overcome these challenges, GoML designed a 5-week Proof of Concept (POC) leveraging AWS infrastructure. Key components of the solution included:

AI-Driven Data Collection: Developed serverless data collection functions using AWS Lambda to fetch relevant data from GroupSolver’s repository, ensuring data accuracy and relevance.

System Integration: Integrated AI models seamlessly within GroupSolver’s AWS environment, enabling smooth operation within their existing platform.

Deployment: Implemented a streaming API for efficient data flow management, with Python for scripting and automation.

LLM Model Development: Fine-tuned Large Language Models (LLMs) using AWS Bedrock with Claude V3 to analyze data and produce insightful summaries.

Testing and Validation: Conducted comprehensive testing to validate the accuracy and effectiveness of AI-generated insights, ensuring high-quality output.

Architecture

• AWS Cloud Infrastructure: CloudWatch: Monitors and logs application and infrastructure metrics, providing insights to detect performance issues and improve reliability.
• Networking and Security: VPC (Virtual Private Cloud):
  Public Subnet
  Web Interface: Hosts the publicly accessible web application for end-users.
  API Gateway: Acts as the main entry point for HTTP requests, routing incoming traffic to internal services within the VPC and ensuring secure access control.
  Private Subnet: Contains backend services that are not directly accessible from the public internet, enhancing security and access control.
• Processing and Data Management:
  Lambda: Executes serverless functions for backend processing, automatically scaling to meet demand while reducing infrastructure management overhead.
  ECR (Elastic Container Registry): Stores Docker container images, enabling seamless deployment of containerized applications and services within the architecture.
  Prompt Engineering Module: Manages and processes input prompts for the AI system to optimize interaction with the language model.
• AI Model and API:
  Anthropic Claude v3: An advanced large language model (LLM) integrated within the architecture, capable of processing natural language prompts generated by the prompt engineering module, handling complex tasks, and generating responses.
  FastAPI: A lightweight and high-performance Python framework used to build RESTful APIs, enabling interaction between different components of the application.
• External Integration and Source Control:
  Bitbucket: Manages source code and integrates with the deployment pipeline, enabling version control, collaboration, and streamlined code deployment.
• Security and Access Control:
  WAF (Web Application Firewall): Protects the application from common web threats like SQL injection and cross-site scripting (XSS) by filtering and monitoring HTTP requests between users and the web interface.