The Singaporean Insurance Pioneer
Overview
THE CHALLENGE
The client struggled with scattered internal systems, slow document retrieval, and needed a secure, scalable, air-gapped GenAI solution
THE SOLUTION
Ollion built a private AWS EKS and Bedrock-based GenAI platform with custom UX, multi-model support, and strict ADFS-enabled security.
THE OUTCOME
The solution boosted productivity, accelerated marketing workflows with image generation, improved employee experience, and scaled efficiently across 360+ users.
About the Business
The Client, a pioneer in the Singaporean insurance sector for over fifty years, aims to make essential insurance accessible to the workforce. Today, as a substantially corporatized entity, it manages assets north of SG$43 billion and caters to more than 1.4 million customers.
A collaboration with Ollion was initiated to introduce a scalable, AI-powered chatbot built on AWS GenAI LLM Chatbot framework and Large Language Models (LLMs), designed to streamline internal operations for around 360 users across Operations, Marketing, IT, and Shared Services. The integration with the client's SharePoint data via Retrieval-Augmented Generation (RAG) has successfully cut down manual workloads and accelerated data retrieval, all while maintaining stringent security and compliance standards.
The business challenges/opportunity
As a major player in the insurance sector, the client manages a vast repository of internal documentation, policies, and procedural data. Employees frequently spent valuable time manually searching through disjointed systems to find relevant information, leading to operational bottlenecks.
The client identified a need for a secure, internal Generative AI solution that could:
Democratize Data Access
Allow users to query internal documents using natural language.
Ensure Data Security
Operate within a secure infrastructure without relying on public internet dependencies for sensitive operations.
Scale Effectively
Support multiple departments with varying needs, from text generation to image creation.
Integrate Seamlessly
Connect directly with existing document repositories (SharePoint) and authentication systems Active Directory Federation Services (ADFS).
The Ollion solution
A unique challenge for this Client, given its D-SII status, is the requirement for air-gapped or strictly private deployments. The solution could not rely on public internet APIs for data processing. Any data sent to an LLM had to stay within a controlled environment, ensuring that customer personally identifiable information (PII) or commercially sensitive underwriting logic was never exposed to the public internet or used to train third-party models. This non-functional requirement eliminated many off-the-shelf SaaS AI tools, necessitating a custom-architected solution on AWS.
The project was executed over an agile 8-week timeline, divided into three focused epics.
1. Architecture & Core Deployment
The foundation of the solution utilizes a containerized architecture on Amazon EKS (Elastic Kubernetes Service) to ensure scalability and ease of management.
- Multi-Model Support: Leveraging Amazon Bedrock, the system was configured to support various LLM providers (e.g., Anthropic Claude, Amazon Titan) allowing the client to switch models based on use-case performance.
- RAG Implementation: To ground the AI's answers in truth, Ollion integrated with the Knowledge Base function within Amazon Bedrock. This allows the chatbot to retrieve context from the client's specific internal documents before generating an answer, significantly reducing hallucinations.
2. Customization & Feature Enhancement
Beyond the standard deployment, Ollion engineered specific customizations to meet the client's functional requirements:
- User Experience (UI/UX): A redesigned interface featuring the client's branding, custom XL banners, and a user-friendly landing page with image associations for different applications.
- Session Management: Implementation of session history visibility, allowing users to revisit previous context and conversations.
- Image Generation Module: A separate, integrated application specifically for generating images (leveraging Amazon Titan and Amazon Nova Canvas) to support the Marketing team.
- Knowledge Base Configuration: A dynamic front-end configuration that lists available knowledge bases, eliminating the need for constant application restarts when data sources change.
3. Security & Compliance
Given the regulated nature of the industry, security was paramount:
- Authentication: Integration with Active Directory Federation Services (ADFS) for secure Single Sign-On (SSO).
- Vulnerability Management: The production deployment underwent rigorous code scanning to ensure it was free of critical and high vulnerabilities.
- Secure Logging: Implementation of encrypted logging to protect sensitive information during interactions.
The Business Benefits
The implementation of the GenAI Chatbot delivered immediate and tangible value to the organization:
The One Big Thing
The singular, transformative outcome of this engagement is the democratization of institutional knowledge.
In a traditional insurance hierarchy, knowledge is hierarchical—senior underwriters know the rules, junior staff ask them. This creates bottlenecks. The Ollion solution flattens this hierarchy. A junior associate in the Operations team can now query the system and receive an answer derived from the same high-level documents used by senior management. This does not replace the senior staff but frees them from answering routine queries, allowing them to focus on complex, high-judgment cases.
CSP involvement
As an AWS Premier Consulting Partner, Ollion leveraged the full depth of the Amazon Web Services (CSP) ecosystem to deliver a secure and future-proof solution:
Amazon Bedrock
Served as the core managed service for accessing foundation models. This serverless approach allowed the client to experiment with multiple top-tier LLMs without the complexity of managing underlying infrastructure or compromising data privacy.
Amazon Kendra
Utilized for its enterprise-grade intelligent search capabilities, providing the "brain" for the Retrieval-Augmented Generation (RAG) system. This ensured that the chatbot's answers were strictly grounded in the client's verified SharePoint data.
Amazon EKS & DynamoDB
Provided the robust backbone for the application. EKS ensured the application container could scale dynamically with user demand, while DynamoDB offered low-latency storage for maintaining conversational context (session memory).
