Uniphore, the global technology company known for its conversational AI and automation solutions, is taking a step towards simplifying how enterprises develop retrieval augmented generation (RAG) applications. The company today announced the launch of X-Stream, a new layer in its core data and AI platform that enables knowledge-as-a-service and brings together powerful tools, connectors and controls for enterprises to mobilize their multimodal datasets for grounded, domain-specific AI applications.
At its core, what X-Stream gives enterprises is a unified and open architecture to combine all the fragmented steps of preparing AI-ready data into a seamless process — essentially serving as a one-stop solution and eliminating the need to use multiple tools across the stack.
“With X-Stream, customers can fine-tune their data, convert it into AI-ready knowledge and seamlessly feed it into Uniphore’s industry-specific, production-ready small language models or build their own. Our data scientists and engineers, drawing on years of experience, have solved for accuracy and hallucinations, ensuring safety and guiding customers towards AI sovereignty,” Umesh Sachdev, the CEO of the company, told VentureBeat.
Solving the data problem for RAG
With the rise of generative AI, the idea of RAG, where AI uses information from a specified set of databases and sources to provide accurate answers to complex questions, has become quite prevalent. Most enterprises today are racing to build dedicated RAG-based search and chat apps that could use their internal knowledge base to provide hallucination-free responses and ultimately drive efficiencies across different functions.
However, when it comes to building (and scaling) such apps, things tend to get a little tricky — especially on the data front.
In almost every case of RAG, the information that an organization wants to use is spread across different sources and formats, from structured tables to unstructured text conversations, documents and videos. To get all this information together, the company has to cobble up several components and use data connectors/ETL tools (like Fivetran) to connect to their respective data warehouses, ERP, HCMs, internal apps etc.
Once the information is connected, they have to enable RAG flow by chunking the data, converting it into embeddings and storing them in a vector database using tools like Milvus, Weaviate or Pinecone. Then, to improve accuracy, they potentially add a graph RAG capability like Neo4j.
All these steps and tools, and then some more, add up very quickly and make it a hard stack to manage and operate. As a result, the project ends up taking months to mature into a scalable gen AI app.
“We have been hearing from enterprise data leaders that they want a more efficient way to drive knowledge transformation from their own data sets across voice, video and text – instead of using traditional data platforms or libraries,” Sachdev said.
To address these data gaps, Uniphore has introduced X-Stream, a unified and open architecture that brings all necessary tools and controls to one place.
The offering ingests multimodal data from over 200 sources and makes it AI-ready by running intelligent merging and transformation jobs. Once the initial processing is complete, it parses and chunks the data, converts it into embeddings and stores them in a vector database, assisting data teams in providing relevant data to AI teams, specifically for feeding Uniphore’s industry-specific small models or their own for RAG and fine-tuning use cases.
But that’s not it.
X-Stream also generates knowledge graphs, where context and reasoning are needed, and creates synthetic data to fine-tune models specific to particular use cases or industries. Plus, it provides evidence management capabilities like factuality checks and chunk attribution to enhance trust in AI.
This essentially gives teams a complete solution to enhance their entire AI pipeline, from data preparation to final output. This allows for the development of production-grade RAG apps much faster.
“X-Stream is distinct for two reasons: it draws from Uniphore’s 16 years of experience working with a variety of unstructured data across voice, video and text, and provides a unified and open platform capability that caters to a broad range of enterprise AI needs,” Sachdev added.
Significant value promised
While X-Stream is new, Sachdev noted that its ability to optimize AI and data components can drive up to 8x faster deployment for domain-specific gen AI apps that use in-house data and meet the highest quality, compliance and governance standards.
“Uniphore offers a usage-based pricing model, and customers typically see a 4x-6x return on investment in weeks from going live,” he noted.
Notably, some of X-Stream’s data capabilities are also provided by hyperscalers and startups, including Amazon (with Sagemaker), Tonic AI and Unstructured.io. It will be interesting how the new offering scales, especially as more enterprises adopt generative AI to power their internal and external use cases. Uniphore works with more than 1,500 companies, including DHL, Accenture and General Insurance.
According to Gartner, throughout 2025, 30% of generative AI projects will be abandoned after proof of concept due to poor data quality, inadequate risk controls or escalating costs.
The post Uniphore unveils X-Stream, a unified knowledge offering to build RAG apps 8x faster appeared first on Venture Beat.
