Title
AWS re:Invent 2022 - Get the most out of your data with ML-powered search (AIM304)
Summary
- Introduction: Jean-Pierre Dodel (JP), Principal Product Manager for Amazon Kendra, introduces the session on ML-powered search with Amazon Kendra, featuring guest speakers Hemant Modi from Qualtrics and Jomo Stark from Orion Health.
- The Problem: Enterprises struggle with 80% of their data being unstructured, leading to frustrations with conventional search technologies.
- Amazon Kendra Solution: Kendra addresses these challenges with natural language queries, state-of-the-art NLU and ML, instant answers, FAQ matching, semantic document ranking, broad domain expertise, and incremental learning for self-improvement.
- Ease of Implementation: Kendra simplifies implementation with connectors for data ingestion, pre-processing mechanisms, and a no-code UI builder for creating search applications.
- Security: Data is encrypted in transit and at rest, with token-based access control and integration with IAM Identity Center for secure user experiences.
- Connectors: Kendra offers over 100 connectors for various data sources, including SharePoint, Confluence, Alfresco, and Box.
- Intelligent Search Experience: Kendra delivers accurate answers quickly, improving the user experience compared to traditional keyword-based searches.
- Tuning Relevance: Kendra allows for relevance tuning with incremental learning and user-friendly controls for adjusting content weighting.
- Use Cases: Kendra is used for enhancing employee experiences with enterprise-wide search applications, transforming customer experiences in call centers and chatbots, and embedding intelligent search in ISV and SaaS applications.
- Getting Started with Kendra: The process involves data ingestion, optimization, and deployment, with options for a no-code UI builder and API integration.
- Customer Stories:
- Qualtrics: Hemant Modi discusses how Qualtrics uses Kendra to streamline support workflows and improve contact center agent experiences.
- Orion Health: Jomo Stark shares how Orion Health integrates Kendra into their digital front door platform to improve healthcare access and information delivery.
- New Features: Kendra introduces tabular search and expands semantic search support for seven additional languages.
- Getting Started: JP recommends running a Flash POC with Kendra to assess its capabilities with your own data.
Insights
- ML-Powered Search: Amazon Kendra's use of machine learning and natural language understanding significantly enhances the search experience by understanding the nuances of language and providing precise answers from unstructured data.
- Domain Expertise: Kendra's pre-training across 14 domains allows for high accuracy without the need for extensive tuning, which is particularly beneficial for enterprises with diverse data sets.
- Incremental Learning: The feature that allows Kendra to learn from user interactions and feedback is a key differentiator, as it enables the search engine to continuously improve its accuracy and relevance over time.
- Customer Adoption: The experiences shared by Qualtrics and Orion Health demonstrate Kendra's versatility and impact in real-world applications, particularly in improving employee productivity and customer satisfaction.
- Rapid Prototyping: Kendra's rapid prototyping capability, as highlighted by Orion Health's six-week prototype development, shows its potential for quick integration and testing, which can accelerate innovation and deployment.
- Tabular Search: The new tabular search feature expands Kendra's capabilities to extract and present answers from HTML tables, which is a common format for structured information on the web.
- Language Expansion: The expansion of semantic search support to additional languages opens up Kendra's benefits to a wider global audience and reflects AWS's commitment to catering to diverse customer needs.
- Ease of Use: The no-code UI builder and API integration options make Kendra accessible to organizations without extensive machine learning expertise, lowering the barrier to entry for adopting advanced search technologies.