![case study](https://cdn.quantiphi.com/2024/06/Milliman-Non-Public-Case-Study-Banner-1.jpg)
Business Impacts
60%
Enhancement in task automation
75%
Improvement in summarization
Customer Key Facts
- Country : US
- Industry : BFSI
Problem Context
The customer, an international actuarial and consulting firm based in Seattle, aimed to streamline their insurance policy drafting process for efficiency.
The customer partnered with Quantiphi to develop a retrieval-augmented generation (RAG) pipeline, customizing it for their insurance policy drafting needs. This pipeline facilitates the identification and synthesis of pertinent legal data from state statutes, integrating input from the firm’s legal team.
Challenges
- Filing process usually takes 6 to 12 months
- Involves multiple review cycles between the customer and regulatory body
![](https://cdn.quantiphi.com/2024/05/Milliman-Non-Public-Case-Study-Challenges-Image.jpg)
Technologies Used
![Vertex AI](https://cdn.quantiphi.com/2024/05/8119a4c3-vertexai-e1656078086129.png)
Vertex AI
![Firebase](https://cdn.quantiphi.com/2023/04/60a58738-firebase.png)
Firebase
![Cloud Run](https://cdn.quantiphi.com/2024/05/8773705b-cloud-run-updated-logo.png)
Cloud Run
![Cloud Storage](https://cdn.quantiphi.com/2024/05/321aa23a-cloud-storage.png)
Cloud Storage
![Vector Search](https://cdn.quantiphi.com/2024/05/8119a4c3-vertexai-e1656078086129.png)
Vector Search
![Cloud Function](https://cdn.quantiphi.com/2024/05/276f58a8-cloud-functions-updated-logo.png)
Cloud Function
Solution
- Developed an LLM-based chatbot to address insurance-related queries for actuaries associated with the customer
- Established a continuous integration and continuous deployment (CI/CD) pipeline to automate data ingestion
- Improved user interaction through functionalities like multi-state response generation, user authentication, and a metrics analysis dashboard
Results
- The LLM-based AI assistant tracks user chat history, allowing easy access during sessions
- The admin page displays KPIs for cost and query count, and lets users save feedback on responses