AWS Elastic Beanstalk Environment Optimization
About The Customer
A digital agency helping the customer harness the power of generative AI for assistance with Human Resources.
Problem Statement
The HR team frequently received repetitive queries regarding policies, benefits, onboarding procedures, and compliance. Managing these inquiries manually was time-consuming and inefficient. The client aimed to automate these responses while ensuring accuracy, reducing HR workload, and improving employee experience.
To address the challenge, an AWS-based Gen AI RAG solution was designed and implemented. The architecture comprised several key components:
1. User Interaction Layer
-
Clients: Employees could access the chatbot via web and mobile interfaces.
-
Chatbot: The chatbot acted as the primary interface, processing user inputs and generating responses.
2. Backend Processing
-
Amazon API Gateway: Managed API requests from the chatbot to backend services.
-
Containerized Backend API: A scalable API deployed on AWS, responsible for handling RAG-based queries and orchestrating interactions between various components.
-
AWS Batch: Handled batch processing tasks, including document processing and embedding generation.
-
Amazon SageMaker: Hosted and managed the machine learning models used for text processing and response generation.
3. Data Processing and Storage
-
S3 (Uploaded Files): HR documents were uploaded and stored securely.
-
AWS Textract: Extracted text from uploaded HR documents.
-
Translation Model: Translated extracted text where necessary for multilingual support.
-
Embedding Model: Converted textual content into vector embeddings for semantic search.
-
RDS PostgreSQL (pgVector): Stored and managed vectorized document embeddings, enabling efficient retrieval of relevant content based on user queries.
4. AI-Powered Response Generation
-
Embedding Model: Transformed user queries into embeddings for similarity search.
-
RDS Search: Retrieved the most relevant content from the document store based on embeddings.
-
LLM (Large Language Model) on SageMaker: Generated responses by combining retrieved document snippets with generative AI capabilities.
Success Factors
-
Enhanced Efficiency: Automated responses reduced HR workload significantly.
-
Improved Employee Experience: Employees received instant, accurate answers to their queries.
-
Scalability: The AWS-based architecture ensured seamless scaling as demand increased.
-
Security & Compliance: The solution adhered to data protection policies, ensuring secure document handling.
-
Cost Optimization: Leveraging AWS services such as S3, Batch, and RDS allowed for optimized costs while maintaining performance.
