Aryn’s new open source, end-to-end conversational search stack makes it easy for developers to build conversational search applications on enterprise data using generative AI, OpenSearch, and semantic data preparation.

Aryn released a new open source conversational search stack for unstructured enterprise data such as documents, presentations, and internal websites. This stack makes it easy for developers to build applications such as question-answering over knowledge bases, chatbots for customer support or documentation, and research and discovery platforms across internal datasets. With conversational search, users can ask natural language questions on large datasets, get high-quality answers, and submit follow-up questions in the context of the interaction. Aryn’s stack is an integrated one-stop shop from data ingestion to conversational APIs, enabling developers to easily build conversational search applications and deploy them to production in days instead of months. It includes OpenSearch, a popular enterprise-grade search engine used by tens of thousands of developers. Aryn adds conversational capabilities to OpenSearch that use generative AI to capture query semantics and synthesize responses from the most relevant sections of the dataset. These generative AI models need high quality data to produce high quality answers. To enable this, Aryn released Sycamore, a new scalable semantic data preparation system purpose-built for search. As part of the stack, Sycamore uses generative AI to clean, enrich, and capture the meaning of data to enable high quality answers. Aryn’s conversational search stack is 100% open source (Apache License v2.0), which gives developers the freedom to customize and build without “vendor lock-in.”

“Aryn’s open source conversational search stack is an integrated one-stop shop that developers can use to easily build conversational applications on unstructured enterprise data and quickly scale them to production.”

With the recent advances in generative AI, enterprises across industry verticals see the potential to unlock the value of their unstructured data through conversational applications. Unfortunately, developers at these companies struggle to build high-quality conversational search apps. They often start with large language models (LLMs), a type of generative AI, which have a remarkable ability to process and generate natural language. However, most LLMs are trained only on public data, and therefore cannot answer questions from private data. They are also notorious for hallucinating incorrect answers. Training these models on private data is expensive and infeasible for many companies. It requires unique expertise, massive amounts of training data, costly hardware, and frequent retraining. Instead of model training, developers can use a technique called retrieval-augmented generation (RAG) to extend LLMs to cover private datasets. The RAG architecture retrieves relevant sections of private data and feeds them to an LLM to constrain the response and generate high-quality answers. Today, RAG requires assembling and scaling complex pipelines of disparate components – from generative AI toolkits to vector databases – that are in their infancy, lack enterprise features like security, and are hard to maintain. Also, generative AI answer quality is only as good as the data it is fed. For example, stale, dirty, or raw data will return subpar results. While developers can use proprietary conversational search solutions, such solutions lack flexibility and lock in customers. These developers want an open source stack that they can use to build high-quality conversational apps.

Aryn’s open source conversational search stack is an integrated one-stop shop that developers can use to easily build conversational applications on unstructured enterprise data and quickly scale them to production. The stack consists of three main components: semantic data preparation with Sycamore, semantic search with OpenSearch, and conversational capabilities in OpenSearch. Generative AI augments each of these components, leading to higher quality answers and ease of use. Developers can easily choose and experiment with different generative AI models and customize their prompts, without needing to become experts in AI and search.

Marketing Technology News: Cisco to Acquire Splunk, to Help Make Organizations More Secure and Resilient in an AI-Powered World

Aryn created and open sourced Sycamore, a robust and scalable semantic data preparation system for unlocking the meaning of unstructured data and preparing it for search. It provides a high-level API to construct Python-native data pipelines with operations such as data cleaning, information extraction, enrichment, summarization, and generation of vector embeddings that encapsulate the semantics of data. Sycamore uses generative AI to make these operations simple and effective. Additionally, Sycamore is a data parallel system, so developers can seamlessly scale data processing and quickly load vector embeddings into OpenSearch.

Aryn’s stack includes OpenSearch for information retrieval on private data. OpenSearch is a popular search engine with a robust set of features, enterprise-grade security, and a growing open source community. Today, thousands of enterprise customers such as Goldman Sachs, Pinterest, and Zoom use OpenSearch. Aryn added support for conversation memory and APIs in OpenSearch v2.10, so that developers can build conversational apps without needing to stitch together and manage third-party tools and libraries. This functionality makes it easy for applications to orchestrate interactions with LLMs, store the history, and use it as context for future interactions. Powering these conversational features is OpenSearch’s rich set of enterprise-grade functionality needed to build robust search applications. The OpenSearch engine includes a vector database and term-indexing functionality to implement multiple search techniques for better relevance. Additionally, OpenSearch is battle-tested and includes enterprise-grade security, scalability, monitoring, and availability. Developers can get applications to production faster with this underlying tried-and-true search platform.

Aryn’s conversational search stack is 100% open source under the Apache v2.0 license, which makes it easy to customize when required and avoids vendor lock-in.

“Over 90% of unstructured enterprise data is untapped because companies don’t have simple tools to process and query that data in natural language and get high quality results,” said Mehul Shah, Co-founder and CEO of Aryn. “Many companies invest significant engineering effort or pay for expensive consultants to build and tune platforms for data preparation and search. Even after all that effort, these platforms often return poor quality results and cannot handle natural language questions. With Aryn’s conversational search stack, developers can easily build chat-style interfaces for their enterprise data. Aryn simplifies data preparation, search, and generative AI, enabling developers to build search applications with high-quality answers and to get to production faster for their customers.”

“Businesses choose OpenSearch because it is a secure, high-quality, open source search and analytics suite with a rich roadmap of new and innovative functionality. The OpenSearch community is excited about how generative AI and vector search is enabling better answer quality and natural language conversations with data,” said Mukul Karnik, General Manager of OpenSearch at Amazon Web Services (AWS). “We are thrilled that Aryn has contributed conversational search and conversational memory functionality to help users to search their unstructured data using natural language and build generative AI chat experiences. Furthermore, we are excited to see how the community uses Sycamore for data preparation and enrichment of unstructured data for conversational search applications.”

Fixie is a platform for building conversational AI Sidekicks that are designed to answer questions, undertake action, and live directly alongside your application. “Many of our customers need their Sidekicks to provide high-quality answers for questions on their data,” said Matt Welsh, Chief Architect and Co-Founder of Fixie. “To provide this capability, we’re partnering with Aryn. Aryn’s stack gives us a one-stop shop for building question-answering over knowledge bases for our customers’ unstructured datasets, along with well-integrated components for all aspects of natural language search. This makes it easy for us to get started and implement conversational search in our service.”

Aryn’s mission is to answer questions from all of your data. To achieve this, Aryn is bringing generative AI to OpenSearch. Also, the company is dedicated to growing and expanding the OpenSearch open source community. Aryn is comprised of data and cloud experts from AWS and Google Cloud with a track record of building and scaling big data cloud services. They have a $7.5M seed investment from Factory HQ, 8VC, Lip-Bu Tan, Amarjit Gill, and other notable angels and advisors.

Marketing Technology News: MarTech Interview with Gary Sabin, VP of Product Management at Impartner