Highlights –

  • Today’s market is flooded with various companies selling vector-database technology like Pinecone. Vectara offers various services, and a vector database is just one of them.
  • Cross-attentional ranking is a method used by the Vectara platform to rate results that consider both the meaning of the query and the outcomes to get even better outcomes.

Is there a more effective way to create a search tool that delivers more highly relevant results than only employing keyword-based strategies?

Amr Awadallah (CEO), Amin Ahmad (CTO), and Tallat Shafaat (Chief Architect), all former Google employees, set out to answer this question and other concerns with their new company, ZIR AI, which had been in stealth mode. ZIR AI is now coming out of stealth under the name Vectara with the aid of USD 20 million in seed funding and the availability of the company’s neural search-as-a-service technology.

Vectara’s core idea is that Artificial Intelligence (AI)-based Large Language Models (LLMs), along with Natural Language Processing (NLP), data integration pipelines, and vector techniques, can generate a neural network with numerous applications, including search.

Awadallah said, “At the heart of what we have built is a neural network that makes it very simple for any company to tap that power and do something useful with it. Large language models and neural networks have transformed how we understand the meaning behind the text, and the first offering we’re launching is neural search-as-a-service.

How Vectara is combining many AI approaches into something new

Vectors serve as a fundamental building block for LLMs and neural networks. “One of the key elements of making large language models and neural network inference is a vector-matching system in the middle,” Awadallah said.

According to Awadallah, neural networks’ input data and the output of the network are the vectors that represent the learnings produced by the network. He emphasized that the Vectara platform spans the entire data pipeline and goes beyond just examining vectors.

Today’s market is flooded with various companies selling vector-database technology like Pinecone. Vectara offers various services, and a vector database is just one of them.

According to Awadallah, when a user raises a query, Vectara utilizes its neural network to translate the request from the language space — which includes vocabulary and grammar—into the vector space, which consists of numbers and math. Vectara then indexes all the information a company wishes to search in a vector database, identifying the vector most closely related to that user query.

A substantial data pipeline that ingests diverse data types feeds the vector database. For instance, the data pipeline can comprehend the structure of PDF and ordinary Word documents and knows how to handle both formats. The Vectara platform makes use of a cross-attentional approach to provide results. It considers both the meaning of the query and the outcomes to get even better outcomes.

Going from big data on Hadoop to neural search-as-a-service

Vectara is not the only startup Awadallah has helped launch. Awadallah also cofounded the Hadoop service company Cloudera in 2008. Lessons from his experiences are being used to guide decision-making at the new firm.

One of the things he has learned over the years is that creating technology only for its own sake is never a good idea. Vectara’s neural data processing pipeline is strong and has a wide range of potential applications, according to Awadallah. At Vectara, they started with a search because it’s a problem that many firms have.

“We wanted to start with a problem that everybody has that needs to be solved in a good way,” Awadallah added.

Awadallah and other cofounders have had prior experience working at Google, a company that used transformer approaches and LLMs. He noted that a transformer makes it feasible to comprehend context more thoroughly to improve a query’s outcome. With the help of a transformer, a firm can not only understand the meaning of a word but also comprehend how the word is linked to other words in that sentence and the previous ones and then the following sentence to get the right context.

He stated, “We did this at Google. We know how to properly fine-tune the parameters to get the best outcome for our customers, and that’s truly what differentiates us.”

Vectara doesn’t just offer search as its first offering. According to Awadallah, his business will gradually add more services, with tools to assist customers in surfacing related topics and making recommendations.

“The Industrial Revolution was about how we make stuff with our hands, and now we’re helping people to build things with stuff that is coming out of their brains. That’s the foundation of this pipeline that we’re building, which is a neural network pipeline that allows you to process and extract value out of data,” Awadallah added.