Jina.ai has received $30 million in funding for its neural search platform

The goal behind neural search, according to Jina.ai CEO and co-founder Han Xiao, who co-founded the firm with Nan Wang and Bing He, is to leverage deep learning neural networks to go beyond standard keyword-based search tools. The company’s main Jina framework can let developers quickly design search tools for their individual use cases by utilising relatively new machine learning algorithms such as transfer learning and representation learning.

“We first employ deep neural networks to transform this data format into a universal representation given an image, audio, video, or whatever,” Xiao added. “It’s mostly a mathematical vector in this situation – 100-dimensional vectors.” The matching [algorithm] then counts the mathematical distance, or vector distance, between these two vectors, rather than the number of letters that match. In this way, you may practically utilise this technology to tackle any type of data search or relevance problem.”

Jina is similar to TensorFlow for search, according to Xiao (TensorFlow is Google’s open-source machine learning framework). Jina intends to define how people construct neural search systems — and become the de facto standard in the process, like how TensorFlow or PyTorch defined how people develop AI systems.

However, Jina is just one of the company’s current offerings. It also includes the Jina Hub, a marketplace where developers can share and find building blocks for Jina-based neural search apps, as well as the Finetuner, a tool for fine-tuning any deep neural network.

“We invested a lot of time and effort over the previous 18 months constructing the basic infrastructure, the base of this massive neural search tower — and that portion is already done,” Xiao explained. “Right now, we’re slowly constructing the first and second floors of this massive structure, and we’re attempting to deliver an end-to-end construction experience.”