OpenSearchCon India 2025

Retrieval-Augmented Generation (RAG) has emerged as a powerful paradigm for grounding large language models (LLMs) in external knowledge, enabling more accurate and contextually relevant responses. However, traditional RAG pipelines often operate in a static, pre-defined manner, limiting their ability to adapt to complex queries and dynamically explore information spaces. This talk introduces "Multimodal Agentic RAG", an innovative approach that integrates autonomous agents into the RAG framework to enhance its reasoning and retrieval capabilities. This talk will delve into the architecture and implementation of multimodal RAG with agentic workflows, showcasing its potential to address limitations of traditional RAG. The architecture is based on OpenSearch's multimodal capabilities, DeepSeek-R1's reasoning capabilities, and custom-tuned embedding models. Furthermore, we will discuss the challenges and future directions of Agentic RAG, including the development of robust planning mechanisms, efficient knowledge source management, and effective agent coordination.

With the rise of Agentic AI, search is no longer just about retrieving information—it’s about enabling AI-driven reasoning, contextual awareness, and decision-making. As AI applications become more autonomous, the need for efficient, scalable, and intelligent search platforms has never been greater.

This session explores how OpenSearch can be leveraged as a foundational platform for Agentic AI applications, enabling capabilities like semantic search, vector-based retrieval, hybrid search, and retrieval-augmented generation (RAG). We’ll discuss architectural patterns, indexing strategies, multi-modal search, and real-time data enrichment, ensuring AI agents can access and utilize knowledge effectively.

By adopting a platform-centric approach, organizations can standardize and scale OpenSearch to serve diverse AI-driven use cases—from chatbots and virtual assistants to complex enterprise automation systems. Join us to discover how search is evolving into a key enabler for the next generation of AI applications.

This session explores how Gaia’s decentralized AI platform integrates with OpenSearch to create scalable, privacy-preserving solutions for semantic search, generative AI applications, and real-time analytics.

In this session, I intend to demonstrate a use case where Gaia’s federated learning models enhance OpenSearch’s vector database capabilities, enabling dynamic personalization and efficient retrieval-augmented generation (RAG) workflows.

Gaia’s AI Framework: How Gaia’s decentralized architecture complements OpenSearch’s ecosystem, particularly in scenarios requiring data privacy and distributed model training (e.g., healthcare or financial analytics).

Demo a RAG Pipeline: Build a real-world demo/example using Gaia’s language models to generate embeddings stored in OpenSearch’s binary vector indexes (new in v2.19), coupled with OpenSearch Flow for automated pipeline configuration.

Community Impact: A look at how this integration supports OpenSearch’s goal of becoming the preferred backend for generative AI while adhering to open governance principles.

Struggling to tame the latency of your OpenSearch query? Atlassian was facing the exact problem with a query which had grown organically over a number of years of relevance improvements. Learn how they dove into the internals of OpenSearch to discover how their query worked, reducing their query latency by 70% in the process.

This session will cover:

* Techniques for identifying problem areas
* Shard topology
* Conjunctive and disjunctive boolean queries and how they impact latency
* track_total_hits, how it works and subtleties of when it doesn't
* Concurrent segment search
* Joins, when denormalisation is faster and when it isn't
* Index sorting
* initial query scoring vs rescoring
* Cross field match vs copy_to
* Parallel recall paths using hybrid queries or multi-search
* Query cache vs pre-computing
* LTR plugin, feature logging and ranking and impact on latency

In this session, Uber, a premier member of the OpenSearch Foundation, will share how the unique demands of their low-latency, high-throughput search platform have driven impactful contributions to the OpenSearch project.

The discussion will explore Uber’s collaboration with the OpenSearch community to integrate critical learnings and innovations from their own search platform. Key highlights include the development of a modular, cloud-native architecture with clear separation between readers and writers, a high-performance messaging protocol for inter-service communication, and a pull-based indexing model.

These advancements, along with contributions from the broader OpenSearch community, are shaping OpenSearch into a cutting-edge, serverless open-source search engine tailored for enterprise needs. Join this session to discover how pioneering work from Uber and OpenSearch is paving the way for the next generation of high performance and scalable search technology in open source.

OpenSearch is a powerful tool for log storage, analysis, and dashboard reporting. In the context of Retrieval-Augmented Generation (RAG) applications with Large Language Models (LLMs), OpenSearch serves as a vital component, acting as a vector database. This presentation explores an architecture and process flow to integrate graph-based context with OpenSearch, enhancing the correlatability of data and improving the context and performance of search results.

Through experimentation with synthetic data from a travel app, we’ll demonstrate how customer activity data—such as travel points, preferences, and interactions with destinations—can be integrated into OpenSearch. By introducing graph points and correlating them with other data (e.g., maps and historical data), we can provide personalized, more relevant search results for users. Additionally, we’ll compare metrics before and after the introduction of graph context to showcase how it improves search performance and the overall user experience in RAG-based systems.