TLDR: Nippon India Mutual Fund has significantly improved the accuracy of its AI assistant responses by implementing advanced Retrieval Augmented Generation (RAG) methods on Amazon Bedrock. This innovative approach, which includes query reformulation and response reranking, has led to over 95% accuracy improvement and a 90-95% reduction in hallucinations, drastically cutting report generation time.
Nippon Life India Asset Management Limited has achieved a breakthrough in the accuracy of its AI-powered assistant responses through the adoption of sophisticated Retrieval Augmented Generation (RAG) techniques on Amazon Bedrock. This strategic implementation addresses the challenges of providing precise and reliable information from extensive enterprise data, a critical requirement for financial services.
The core of Nippon India Mutual Fund’s enhanced solution lies in its advanced RAG methods, which go beyond standard approaches. The company has integrated query reformulation and response reranking to optimize the relevance of retrieved information. Specifically, the system converts user queries into embeddings using the Amazon Bedrock embedding model. It then employs a multi-query RAG strategy, generating multiple variants of a user’s query and executing them in parallel. The results from these parallel queries are subsequently reranked using Amazon Bedrock’s reranker models, which calculate the relevance of data chunks and reorder them to ensure the most pertinent information is prioritized. This process ensures that even when a direct similarity match might not place the most relevant data at the top, the reranking mechanism brings it to the forefront.
This advanced RAG implementation has yielded impressive results for Nippon India Mutual Fund. The accuracy of AI assistant responses has seen an improvement of over 95%. Concurrently, the rate of “hallucinations” – instances where the AI generates incorrect or fabricated information – has been reduced by a remarkable 90-95%. Furthermore, the solution now allows for the inclusion of source chunks and file links through file metadata, significantly boosting user confidence in the provided responses. A notable operational efficiency gain is the reduction in report generation time, which has plummeted from approximately two days to just about 10 minutes.
Also Read:
- AI Tools Boost Financial Data Analysis Speed and Accuracy
- Accelerating Drug Discovery: Building AI Research Assistants with Strands Agents and Amazon Bedrock
This initiative underscores the growing importance of robust AI solutions in the financial sector, where accurate and timely information is paramount. By leveraging Amazon Bedrock’s capabilities, including its choice of high-performing foundation models and managed RAG experience through Knowledge Bases, Nippon India Mutual Fund has set a new benchmark for AI assistant performance in enterprise environments. The collaboration with Amazon Bedrock, as highlighted by co-written insights from Abhinav Pandey of Nippon Life India Asset Management Ltd., demonstrates a commitment to continuous improvement in generative AI applications for critical business functions.
