Inside Nishkarsh Srivastava’s Journey to Build Cortex, the Intelligent Retrieval Layer for AI Applications

Knowledge workers are using more applications across their workflows than ever before. Whether it’s Slack for internal comms, Notion for knowledge base, or Jira for project management—not to mention good-ol’ email. 

While these systems have traditionally been siloed from one-another, AI promises the opportunity to comprehend data across systems so that users can access the information they need in seconds. But many developers building AI search and agents have run into the same roadblock. Every application has a different modality–different forms of text, different information structures—which makes it difficult for developers to process them together. 

After encountering this himself when working on a research project at Stanford, Nishkarsh realized that building a vector database that could help developers digest context from different applications at scale would be a bigger problem to focus on. So he set out to build Cortex.

Introducing Cortex: The Retrieval Layer for AI-Native Applications

Cortex is the world’s first vector database that can understand context from external applications like Gmail, Slack, and Notion. 

Unlike traditional search, which relies on keyword matching, AI-powered search processes information based on context, intent, and relationships across different types of data. And moving forward, AI agents searching for context will overshadow any number of searches we’ve done previously. 

But as many popular tools have different modalities—Slack has an individual message structure, Notion has a blog-based structure, and Jira has a ticketing structure—AI apps must be able to understand the context of information in these different formats in order to deliver an effective solution. 

To enable this using the vector databases available, developers must build and maintain their own embedding pipelines, scale them in production, and write their own ingestion and parsing engines. As a result, developers are increasingly spending more time on figuring out why retrievals aren’t working than building actual AI products. And many fail to design effective solutions—MIT recently reported that 95% of Gen AI startups fail their pilots with enterprises, and the number one reason is because their AI is unable to digest context. “The average developer doesn’t want to spend time on these things, or even know how to do these things,” Nishkarsh explains. “And they shouldn’t have to care about them.” 

Instead, Cortex serves as an adaptive retrieval layer that can process information from various data sources with unique modalities, offering context-aware intelligence that works out of the box. This gives AI the fastest, highest quality, and most comprehensive search, and allows developers to spend more time building and less time managing brittle infrastructure.

Since launching just six months ago, Cortex is already being used by several of the Bay Area’s fastest-growing Series A startups as well as publicly-listed companies.

Nishkarsh’s journey to Cortex

Nishkarsh first discovered the problem space when working on a research project at Stanford. He and his team were trying to build a consumer application that would serve as a magical search experience across apps like email, Slack, and OneDrive. “Like ChatGPT, but connected to all of your workbase applications,” he explains. 

They soon realized, however, that there was a bigger problem to solve. The existing vector database solutions they were trying to use were incomplete, clunky, and frustrating. The larger problem to focus on, they believed, would be to help developers digest context from different applications, scale their pipelines in production, optimize search in real time, and ultimately build recommendation engines without having to write a single line of code. 

Nishkarsh and his team started by building a solution for their consumer app, Findr. And after Findr found strong initial traction, ranking as runner-up in ProductHunt’s 2024 Golden Kitty Awards for personal productivity acts, the team expanded their focus to making Cortex available for other developers as well. 

In the long run, Nishkarsh is aiming for Cortex to become a superbase for AI retrievals, allowing any person who wants to become an AI developer to build the best AI products from their laptops without any help.

Nishkarsh’s advice for fellow founders

Although technical capabilities are a big part of Cortex’s differentiation, Nishkarsh explains that it was actually a GTM shift that unlocked growth opportunities for the company.

“You can have the world’s best product, but if you don’t know how to communicate your benefits, no one is going to use it,” he explains. Cortex initially positioned the tool as a solution for improving search, which confused potential customers as many were already using vector databases that offered an internal search feature. When they changed their messaging to communicate that they were actually a better vector database, they were able to convert people from their competitors to their offering. 

“It’s been said so much that distribution is important. But how you distribute is equally important. I could have spent a million dollars on ads saying that Cortex helps build better search, but it wouldn’t work if I didn’t use the right words. Using the right messaging is very important.” 

To learn more about Nishkarsh’s journey, you can follow him on LinkedIn here. And if you’re building an AI application and are looking for a vector database, you can learn more about Cortex at usecortex.ai/.