Conversational Chat Agent for Doppelio

Business Problem

  • Their current OpenAI implementation struggled with consistency in responses, particularly when processing domain-specific documents and testing data
  • Inefficient document processing and information retrieval, requiring manual intervention to extract and analyze data from various document formats
  • The existing solution couldn’t reliably handle the growing number of concurrent enterprise users, leading to degraded performance during peak usage
  • There was a strong need for a secure, scalable, multi-tenant AI solution capable of supporting concurrent users across large enterprise teams
  • Requirement for a secure, multi-tenant architecture that respects data privacy and segregation

About

Doppelio is a leading provider of AI-driven testing and automation solutions that help enterprises validate and optimize their digital experiences. As part of their AI expansion strategy, Doppelio aims to develop a scalable conversational chat agent that integrates seamlessly into enterprise workflows. This solution is designed to provide context-aware, real-time responses tailored to specific business needs, enhancing operational efficiency and delivering intelligent, data-driven insights to users.

Explore Now

Solution

goML developed a Conversational Chat Agent MVP leveraging Generative AI capabilities and a robust AWS-native architecture. A key part of the engagement involved migrating Doppelio’s existing OpenAI-powered solution to AWS Bedrock, significantly improving cost efficiency, latency, and accuracy while aligning with enterprise cloud strategy.

1.

OpenAI to AWS Bedrock Migration
Doppelio’s existing OpenAI integration was migrated to AWS Bedrock, resulting in enhanced control, reduced latency, and improved model accuracy for domain-specific tasks.

4.

Multi-modal Interface
Support for both text and voice-based interactions with improved accuracy and response times

7.

Scalable Infrastructure
Powered by FastAPI and Python, capable of handling 1,000+ concurrent user sessions with low-latency performance.

2.

Natural Language Interface
Built using Claude 3.5 via AWS Bedrock, supporting both text and voice interactions.

5.

Document Intelligence
Enabled parsing and understanding of PDFs, Word docs, and images for test-related insights

3.

Enhanced RAG Implementation
Built a more sophisticated Retrieval-Augmented Generation system specifically optimized for document processing and contextual understanding

6.

Multi-Tenant Secure Architecture
Ensured strict data segregation and encryption using AWS S3, DynamoDB, Lambda, CloudWatch, and API Gateway.

Architecture

  • Input Layer
    • Accepts user inputs in multiple formats: text messages, voice commands, and document uploads (PDF, Word)
    • Provides a unified entry point for all user interactions regardless of format
    • Handles initial preprocessing of data before passing to infrastructure layer
  • Infrastructure Layer (AWS Services)
    • AWS S3: Stores uploaded documents and processed data securely with improved encryption compared to previous solution
    • AWS Lambda: Executes serverless functions with predictable performance characteristics and auto-scaling
    • API Gateway: Manages API endpoints with enhanced throttling controls to prevent performance degradation
    • DynamoDB: Maintains multi-tenant data with proper segregation and enterprise-grade security
    • CloudWatch: Provides comprehensive monitoring and auto-scaling triggers based on actual usage patterns
  • Application Layer
    • Conversational Chat Agent API: Built with FastAPI in Python for superior performance compared to previous implementation
    • Enhanced RAG Processing: Optimized Retrieval-Augmented Generation specifically tuned for Claude 3.5 to improve response accuracy
    • Guardrails: Enterprise-specific content filtering and security controls
  • Intelligence Layer
    • AWS Bedrock:     Provides managed foundation model access with better performance characteristics than OpenAI
    • Claude 3.5: Delivers more accurate and consistent responses, especially for technical document processing
    • User-Imported Models: Supports custom models with better integration than the previous solution
Conversational Chat Agent for Doppelio
Outcomes

Enhanced Document Processing:
RAG capabilities optimized for Claude 3.5 resulted in 67% improvement in information extraction from technical documents.

Improved Accuracy Post-Migration:
The switch from OpenAI to Claude 3.5 via Bedrock led to
higher domain-specific accuracy, better latency, and
streamlined cost structure.

99.95% System Uptime:
Highly stable performance under peak loads
(1,000+ concurrent sessions).

Enterprise Alignment:
Migration to AWS Bedrock aligned with Doppelio’s
cloud-first policy, improving control over model access, security,
and compliance.