Uniphorthe worldwide 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 brand new layer in its core Data and AI platform that allows knowledge as a service and brings together powerful tools, connectors and controls to assist organizations mobilize their multimodal data sets for informed, domain-specific AI applications.
At its core, X-Stream provides enterprises with a unified and open architecture to mix all of the fragmented steps of AI-enabled data preparation into one seamless process – essentially serving as an entire solution and eliminating the necessity to use multiple tools across the stack.
“With X-Stream, customers can optimize their data, transform it into AI-ready knowledge, and seamlessly feed it into Uniphore's industry-specific, production-ready small language models or construct their very own. Our data scientists and engineers have leveraged years of experience to develop solutions for accuracy and illusion to make sure security and guide customers on the trail to AI sovereignty,” Umesh Sachdev, the corporate's CEO, told VentureBeat.
Solving the info problem for RAG
With the appearance of generative AI, the concept of RAG, where AI leverages information from a spread of databases and sources to supply accurate answers to complex questions, has gained traction. Most firms today are working on developing dedicated RAG-based search and chat apps that leverage their internal knowledge base to supply error-free answers, ultimately increasing efficiency in various functions.
However, constructing (and scaling) such apps could be a bit tricky – especially with regards to data.
In just about all cases of RAG, the knowledge a corporation desires to leverage is spread across multiple sources and formats, from structured spreadsheets to unstructured text conversations, documents, and videos. To tie all of this information together, the organization must cobble together multiple components and use data connectors/ETL tools (like Fivetran) to hook up with their respective databases, ERP systems, HCMs, internal apps, etc.
Once the knowledge is linked, they should enable the RAG flow by splitting the info into chunks, turning them into embeddings, and storing them in a vector database using tools like Milvus, Weaviate, or Pinecone. To improve accuracy, they might then add a graph RAG function like Neo4j.
All these steps and tools after which some add up in a short time and make it a difficult stack to administer and use. As a result, it takes months for the project to mature right into a scalable Gen-AI app.
“We’ve heard from enterprise data leaders that they desire a more efficient option to drive knowledge transformation from their very own data sets across voice, video and text – slightly than using traditional data platforms or libraries,” said Sachdev.
To close these data gaps, Uniphore introduced X-Stream, a unified and open architecture that brings together all of the crucial tools and controls in a single place.
The offering ingests multimodal data from over 200 sources and makes it AI-enabled through intelligent merging and transformation jobs. Once the initial processing is complete, the info is analyzed and split into chunks, transformed into embeddings, and stored in a vector database, helping data teams provide relevant data to AI teams, especially to feed Uniphore's industry-specific small models or their very own for RAG and fine-tuning use cases.
But that's not it.
X-Stream also generates knowledge graphs when context and reasoning are required and creates synthetic data to optimize models specifically tailored to certain use cases or industries. It also provides evidence management capabilities resembling fact-checking and chunk mapping to extend trust in AI.
This essentially gives teams an entire solution to enhance their entire AI pipeline, from data preparation to final output, enabling them to develop production-ready RAG apps much faster.
“X-Stream is exclusive for 2 reasons: it leverages Uniphore’s 16 years of experience working with quite a lot of unstructured data across voice, video and text, and it provides a unified and open platform capability that addresses a wide selection of enterprise AI needs,” added Sachdev.
Significant added value promised
Although X-Stream is recent, Sachdev identified that the app's ability to optimize AI and data components can result in as much as eight times faster deployment of new-generation, domain-specific AI apps that use internal data and meet the best quality, compliance and governance standards.
“Uniphore offers a usage-based pricing model and customers typically achieve a 4 to 6 times return on investment inside a couple of weeks of going live,” he noted.
Some of X-Stream’s data capabilities are also provided by hyperscalers and startups, including Amazon (with Sagemaker), AI Tonic And Unstructured.ioIt will probably be interesting to see how the brand new offering scales, especially as increasingly firms adopt generative AI for his or her internal and external use cases. Uniphore works with greater than 1,500 firms, including DHL, Accenture and General Insurance.
Accordingly GardenerBy 2025, 30% of generative AI projects will probably be abandoned after proof of concept attributable to poor data quality, inadequate risk controls, or rising costs.
