Uniphore unveils X-Stream, a unified knowledge offering to build RAG apps 8x faster

Uniphore, the worldwide know-how firm identified for its conversational AI and automation options, is taking a step in the direction of simplifying how enterprises develop retrieval augmented era (RAG) purposes. The corporate immediately introduced the launch of X-Stream, a brand new layer in its core data and AI platform that permits knowledge-as-a-service and brings collectively highly effective instruments, connectors and controls for enterprises to mobilize their multimodal datasets for grounded, domain-specific AI purposes.

At its core, what X-Stream provides enterprises is a unified and open structure to mix all of the fragmented steps of making ready AI-ready information right into a seamless course of — primarily serving as a one-stop answer and eliminating the necessity to use a number of instruments throughout the stack.

“With X-Stream, prospects can fine-tune their information, convert it into AI-ready information and seamlessly feed it into Uniphore’s industry-specific, production-ready small language fashions or construct their very own. Our information scientists and engineers, drawing on years of expertise, have solved for accuracy and hallucinations, making certain security and guiding prospects in the direction of AI sovereignty,” Umesh Sachdev, the CEO of the corporate, advised VentureBeat.

Fixing the information downside for RAG

With the rise of generative AI, the thought of RAG, the place AI makes use of info from a specified set of databases and sources to supply correct solutions to complicated questions, has grow to be fairly prevalent. Most enterprises immediately are racing to construct devoted RAG-based search and chat apps that might use their inner information base to supply hallucination-free responses and in the end drive efficiencies throughout totally different capabilities.

Nonetheless, with regards to constructing (and scaling) such apps, issues are inclined to get just a little tough — particularly on the information entrance. 

In nearly each case of RAG, the data that a company needs to make use of is unfold throughout totally different sources and codecs, from structured tables to unstructured textual content conversations, paperwork and movies. To get all this info collectively, the corporate has to cobble up a number of elements and use information connectors/ETL instruments (like Fivetran) to hook up with their respective information warehouses, ERP, HCMs, inner apps and many others. 

As soon as the data is related, they should allow RAG circulate by chunking the information, changing it into embeddings and storing them in a vector database utilizing instruments like Milvus, Weaviate or Pinecone. Then, to enhance accuracy, they doubtlessly add a graph RAG functionality like Neo4j. 

All these steps and instruments, after which some extra, add up in a short time and make it a tough stack to handle and function. Consequently, the mission finally ends up taking months to mature right into a scalable gen AI app.

“Now we have been listening to from enterprise information leaders that they need a extra environment friendly technique to drive information transformation from their very own information units throughout voice, video and textual content – as a substitute of utilizing conventional information platforms or libraries,” Sachdev stated.

To deal with these information gaps, Uniphore has launched X-Stream, a unified and open structure that brings all essential instruments and controls to 1 place.

The providing ingests multimodal information from over 200 sources and makes it AI-ready by working clever merging and transformation jobs. As soon as the preliminary processing is full, it parses and chunks the information, converts it into embeddings and shops them in a vector database, helping information groups in offering related information to AI groups, particularly for feeding Uniphore’s industry-specific small fashions or their very own for RAG and fine-tuning use instances.

However that’s not it.

X-Stream additionally generates information graphs, the place context and reasoning are wanted, and creates artificial information to fine-tune fashions particular to specific use instances or industries. Plus, it supplies proof administration capabilities like factuality checks and chunk attribution to reinforce belief in AI. 

This primarily provides groups an entire answer to reinforce their complete AI pipeline, from information preparation to closing output. This enables for the event of production-grade RAG apps a lot quicker. 

“X-Stream is distinct for 2 causes: it attracts from Uniphore’s 16 years of expertise working with quite a lot of unstructured information throughout voice, video and textual content, and supplies a unified and open platform functionality that caters to a broad vary of enterprise AI wants,” Sachdev added.

Important worth promised

Whereas X-Stream is new, Sachdev famous that its means to optimize AI and information elements can drive as much as 8x quicker deployment for domain-specific gen AI apps that use in-house information and meet the best high quality, compliance and governance requirements.

“Uniphore gives a usage-based pricing mannequin, and prospects usually see a 4x-6x return on funding in weeks from going dwell,” he famous.

Notably, a few of X-Stream’s information capabilities are additionally offered by hyperscalers and startups, together with Amazon (with Sagemaker), Tonic AI and Unstructured.io. It will likely be fascinating how the brand new providing scales, particularly as extra enterprises undertake generative AI to energy their inner and exterior use instances. Uniphore works with greater than 1,500 firms, together with DHL, Accenture and Common Insurance coverage.

In response to Gartner, all through 2025, 30% of generative AI initiatives will probably be deserted after proof of idea as a consequence of poor information high quality, insufficient danger controls or escalating prices.