Slide 1: Generative AI Technology and Product Update
Generative AI Technology and Product Update July 17, 2023 1
Slide 2: Forward Looking Statements
This presentation and the accompanying oral presentation contain forward-looking statements that involve risks and uncertainties concerning our expectations on the impact of artificial intelligence ("AI") on the technology landscape and the expected strength, performance or benefits of our offerings
Slide 3: Generative AI at Elastic
Generative AI at Elastic Product Update Ken Exner Matt Riley Nick Chow Shay Banon Chief Product Officer GM, Elastic Search Principal Product Manager Founder and Elastic Search Chief Technology Officer
Slide 4: Elasticsearch
103K+ GitHub Stars 148K+ Commits Trusted by 50% of the Fortune 500 Elasticsearch The World’s Most Popular Search Engine Software 4B+ Downloads 6B+ Search requests on Elastic Cloud everyday 3500+ Unique Authors 44
Slide 5: One Data Platform: Out-of-the-box Solutions and the Freedom to Build Anything
One Data Platform Out-of-the-box Solutions And the Freedom to Build Anything Out of the Box Solutions Build Your Own Observability Security Search Elasticsearch Platform Ingest Everything Index, Store, and Analyze Search, Visualize, and Engage 5
Slide 6: Generative AI has initiated a dramatic transformation of the technology landscape
Generative AI has initiated a dramatic transformation of the technology landscape Novel Interaction Model Rising Expectations Emerging Ecosystem Context-aware chat interfaces, Rapid, widespread consumer A new ecosystem of with powerful natural language adoption has transformed technologies, startups
Slide 7: History of AI at Elastic
History of AI at Elastic ELSER - a proprietary Support for NLP and ESRE zero-shot retrieval Support for BYO Transformer Elasticsearch Native support for model supervised and models Relevance Engine Vector Search unsupervised learning Foundational technical building 2017 2019 2022 2023 2023 blocks wi
Slide 8: Elastic provides the bridge between private data and GAI
Elastic provides the bridge between private data and GAI Sensitive databases Multi-system / cloud information Private knowledge bases Case histories 9
Slide 9: Limitations in domain-specific topics can be addressed with various techniques
Limitations in domain-specific topics can be addressed with various techniques Fine Tuning Further training the base LLM with the domain-specific content that is specific to the use case. Expensive to train, expensive to host custom models, and need to be continuous. Retrieval Augmented Generation P
Slide 10: Generative AI
Generative AI What are the key aspects of the company's 401k policy for original question an employee in my location and how do I enroll? ✓ Takes action and creates new content × Generic context × Hallucinations – confident tone, erroneous info 11
Slide 11: Generative AI + Retrieval Augmented Generation
Generative AI + Retrieval Augmented Generation What are the key aspects of original question the company's 401k policy for an employee in my location and how do I enroll? context window Question as search query Most relevant data to ✓ GAI response based on the query the most relevant data Elasticsea
Slide 12: Elasticsearch Relevance Engine
Elasticsearch Relevance Engine™ Gives developers a powerful set of machine learning tools to build AI-powered search applications that integrate with large language models Ability to integrate with Vector database Ability to host your own 3rd party transformer transformer model models (OpenAI) RRF -
Slide 13: Four key investments that add up
Native support for fast similarity search in Vector vector space, alongside the existing Database retrieval models built into Elasticsearch Bring-your-own Create vector embeddings from natural Four key Transformer Model language text directly in your Elasticsearch deployment investments Native that
Slide 14: Vector Database
Vector Database ● Fast similarity search in vector space, alongside the existing retrieval models built into Elasticsearch ● Native implementation of the HNSW algorithm for approximate nearest-neighbor search ● Implemented directly in Lucene, the best, most widely-distributed relevance engine in the
Slide 15: Bring-your-own Transformer Model
Bring-your-own Transformer Model ● Create vector embeddings from natural language text directly in your Elasticsearch deployment ● Fully integrated workflow, avoids syncing between disparate tools and data sources, and captures full computational workload ● Positioned for interoperability with open
Slide 16: Native Transformer Model
Native Transformer Model Elastic's proprietary model, ELSER, improves out-of-the-box ("zero-shot") relevance for any text-based retrieval application Brings state-of-the-art relevance to the entire existing Elasticsearch customer base and community Simple to use and does not require corpus-specific
Slide 17: Integrated Relevance Engine
Integrated Relevance Engine Intuitively integrated into _search API and Reciprocal Rank Fusion (RRF) for best-in-class ranking, filtering, document-level permission enforcement Privacy-first approach w/ native document-level security Hybrid search capabilities combines the best of text- and vector-b
Slide 18: Broad scope of capabilities needed for Generative AI
Broad scope of capabilities needed for Generative AI Most Vector Databases Hybrid Search (text + vector) Store & Search Some Vector Databases Vector Embeddings Choice & Flexibility of embedding models Elasticsearch Filtering & Faceting Create Vector Embeddings Autocomplete Search Optimized for text,
Slide 19: Elasticsearch: production ready search at scale
Elasticsearch: production ready search at scale Deliver at speed Proven at scale Built-in security Resilient Flexible deployment Get up and Wide install base OS Hardening, Enterprise ready, Choose a cloud, running in across industry, network security backed by hybrid, on-prem, minutes, provide geo,
Slide 20: Generative AI with Elastic
Superior relevance Bring your own transformer models into Elastic, integrate with third-party models, or use Elastic’s Learned Sparse Encoder. Generative AI with Elastic Factual, updated, & contextual data Use synced data in Elastic for accurate, up-to-date, business-specific results that pass via c
Slide 21: Generative AI + Retrieval Augmented Generation
Generative AI + Retrieval Augmented Generation What are the key aspects of original question the company's 401k policy for an employee in my location and how do I enroll? context window Question as search query Most relevant data to ✓ GAI response based on the query the most relevant data Elasticsea
Slide 22: In Summary
In Summary ● Elastic provide the bridge between companies’ private data and public LLMs ● Retrieval Augmented Generation (RAG) is the best way to pass context for Gen AI ● RAG requires vector search, relevance modeling, transformer integration, inference APIs, and more ● Elastic provides RAG with su