Redis is much more than a database 🛢

CV Deep Dive

Today, we’re talking with Manvinder Singh, VP of Product Management for AI and Search at Redis.

Redis is a real-time data platform that has long been synonymous with performance and speed in the web and mobile application stack, and is now evolving into a core context engine for AI applications. Built to support low-latency, high-throughput workloads, Redis enables developers to power AI agents, chatbots, search systems, and recommendation engines by providing fast access to structured and unstructured data, vectors, and agent memory. Its goal is to help teams improve accuracy, reduce latency, and scale AI applications efficiently by bringing the most relevant context to LLM-powered systems.

Today, Redis is used by millions of developers worldwide and has been adopted across industries including healthcare, ecommerce, social media, and consumer platforms. Customers use Redis for use cases ranging from RAG-based internal knowledge systems and web-scale search to feature stores and real-time personalization. As the AI stack continues to evolve, Redis has expanded its platform with capabilities like vector and hybrid search, semantic caching, agent memory, and unified feature management to support production-grade AI workloads.

In this conversation, Manvinder shares how Redis is positioning itself as the context engine for the AI stack, the challenges of building in a rapidly evolving ecosystem, and his vision for how agentic systems and context engineering will shape the future of AI-powered applications.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Manvinder 💬

Manvinder, welcome to Cerebral Valley! First off, introduce yourself and give us a bit of background on you and Redis. What led you to join Redis?

Hey there! My name is Manvinder and I’m the VP of Product Management for AI and Search at Redis. I’ve been with the company for about a year and a half and I’m based in San Francisco. 

Prior to Redis, I was at Google for close to 11 years. I joined Redis because I saw a huge opportunity for it to play a role in the AI stack similar to the role it played in the web application and mobile application stack. Redis has been synonymous with performance and speed in the web stack and is used by millions of developers worldwide. 

As this AI revolution is taking place, Redis can play a similar role, and that’s what excited me to join the company.

How would you describe Redis to the uninitiated developer or AI team?

For the AI team, Redis really is the real time context engine for AI agents. What I mean by that is most applications today are being built using LLMs, where LLMs provide reasoning and intelligence as part of the application. But these LLMs don’t act in a vacuum. They operate based on context, which means they need information about the problem they are trying to solve. This has led to an entire field of context engineering being born. 

What Redis is trying to do in the AI stack is be the engine for your context, bringing in diverse sets of data to your agent and helping you find the most relevant information to surface to your applications.

Tell us about your key users today. Who would you say is finding the most value in what you're building with Redis in the AI stack?

Anybody building AI applications using LLMs will find Redis very useful. Our customers typically break down into three segments. There are many developers trying to solve for accuracy. One of the biggest challenges today is relevancy, accuracy, and hallucinations, whatever you want to call it. They need a solution that helps them find the most relevant data to surface to their applications. Those developers look for a search platform and a data platform that can help, and that’s where Redis comes in.

Then there are developers who are further along in their AI journey. They have AI agents and are ready to scale their applications. What they start to see are challenges with scaling, where the cost of using LLMs becomes an issue, or the latency and performance of LLMs becomes a problem, since LLMs are inherently slow. They come to Redis in a similar way developers did in the web era, to bring efficiency into their applications, optimize token usage, and improve speed and latency.

Talk to us about some existing use-cases for Redis. Any interesting customer stories you’d like to highlight? 

In terms of interesting use cases, one big area is AI agents. A lot of customers are using Redis to build AI agents. A great example is Amgen, one of the world’s largest healthcare companies. They built their internal RAG-based application using Redis as the platform. For them, having accurate results for medical researchers searching through their knowledge base was critical, and that was a major reason they chose Redis. This includes teams building chatbots and agents who see Redis as a core data platform.

Another major use case is customers building search at web scale. This could be an ecommerce company or a social media company. What they care about most is very low latency while searching across millions or even hundreds of millions of items. An example is CP Extra, one of the largest companies in Thailand. They built their ecommerce platform using Redis as the search engine powering their entire experience.

Another important use case and customer segment is recommendation systems and MLOps pipelines for recommendation engines. Redis is used as a feature store and as a platform to serve data quickly to machine learning models. A great example is iFood, a major food delivery platform in Latin America. They use Redis as a feature store as well as a vector store for their recommendation and personalization platform.

Walk us through Redis’ platforms. Which use-cases should new customers experiment with first, and how easy is it for them to get started? 

Redis as a data platform has been around for a long time, supporting many use cases focused on delivering performance. When it comes to AI applications specifically, there are four main products we focus on to solve these use cases.

First, Redis has built-in search capabilities through the Redis Query Engine. This is our search product, used for vector search and hybrid search. Customers building chatbots and agents often start here and use it to power their platforms.

Second, we launched a product called Redis LangCache. LangCache is a managed service for semantic caching. The idea is that you can store LLM responses in the cache, and for future queries, instead of calling the LLM again, you check the cache first. This helps optimize token usage and reduce latency. LangCache provides semantic caching out of the box and is very popular with customers running LLMs at scale who want to improve performance.

Third, for customers building agents, we offer a product for managing agentic memory. When you are building agents, you often need to store long-term memory, user preferences, and similar data, and persist that across user sessions. We launched a product called Redis Agent Memory Server, which allows you to extract and serve memories for these applications.

Lastly, for customers building feature stores, we offer a unified feature platform. This includes Redis as the online store, along with Featureform, which is a feature platform you can use to deploy features, manage features, and handle feature governance. It’s used by MLOps teams. So overall, those four products are the Redis Query Engine, LangCache, Agent Memory Server, and the feature platform with Featureform.

How are you measuring the impact and/or results that you’re creating for your key customers? What are you most-heavily focussed on metrics-wise? 

What we look at as a company is obviously adoption of our products, but when we work with customers in each scenario, there is usually a specific metric that matters most to them. For example, if a customer is building a chatbot or an agent, what really matters is the relevance of the responses they get at the end of the application. We measure that using different frameworks that allow you to put metrics around relevance. This includes things like precision and recall. The goal is to develop metrics that align with the use case and give you a way to measure the overall accuracy of the chatbot.

Another example is ecommerce. If you are using Redis as a search engine to power an ecommerce experience, you care about different metrics. You might look at conversion rates, such as how many customers who searched for a product ended up buying it or clicking through to the product page. Those metrics are more aligned with business outcomes, like conversion rates, rather than model accuracy alone.

There are a number of companies working on AI Agent infrastructure. What sets Redis apart from a product or technical perspective?

It really comes down to three things, all of which are part of the DNA and philosophy of Redis. The first is performance. This is true across all of our products. For example, when it comes to vector search, we built the fastest vector search product on the market. That performance delivers low latency and high throughput.

The second is that Redis is seen by developers as a unified platform. Redis has always been very flexible in terms of the data structures and data types it supports. You can use one unified platform for agentic context that stores structured data, unstructured data, and time series data. You don’t need seven different databases for a single application. You can just use Redis as a unified platform for your use cases.

The third is accuracy. I mentioned this in the chatbot example. When you are building AI applications, you are trying to get the most accurate results possible, which means searching for the most relevant data. Redis offers capabilities like hybrid search as part of our vector search solution, along with optimized approaches to quantization and different algorithms that help developers build more accurate applications. It comes down to speed, accuracy, and having a unified platform.

Could you share a little bit about how Redis’ platforms actually work under the hood? What’s the reasoning behind some of the architectural decisions you made?

Speed is part of the DNA of the company and a core value, and we use that every time we make an engineering decision. One example is how we think about performance beyond the fact that Redis is an in-memory database. We have invested heavily in horizontal scaling. As your data grows, we add more shards and aggregate results in a highly optimized way so latency impact is minimal. Scaling while keeping performance front and center is critical.

We also focus on vertical scaling. If you need higher throughput or performance, you can add more vCPUs to get more out of the system. Finally, the algorithms we choose matter. How we do pre-filtering and post-filtering during vector search is optimized for performance. It’s really everything across the stack: horizontal scaling, vertical scaling, algorithm choices, and being in memory. The combination of all of that delivers better performance.

What has been the hardest technical challenge around building Redis into the platforms it is today? 

The hardest and also the most exciting part of building Redis right now is how rapidly the AI stack is evolving. Six months ago, most developers building agents were using LangGraph as their agent framework. Today, there are many more choices. Google, Amazon, and Microsoft have all launched their own frameworks.

For us, as a platform that provides vector search, agent memory, and semantic caching, we need to integrate with all of these frameworks. That’s why we have a very dedicated effort to make sure Redis works well across the ecosystem and integrates cleanly with these tools. More broadly, for anyone building applications today, the biggest challenge is that the stack keeps evolving, with new technologies and frameworks emerging constantly.

How do you foresee Redis evolving over the next 6-12 months? Any product developments that your key users should be most excited about? 

What I’m most excited about is the evolution of the context engine. So far, the way developers have been surfacing data to AI applications has primarily been through search queries, where you do mathematical calculations to find the most relevant data. With LLMs and agents, these agents can now look through data and find what’s most relevant in a way that’s much closer to how a human would.

That’s why we’ve started to see new protocols like MCP or A2A. These protocols surface your tools and data as resources to the agent and then rely on agentic reasoning to determine what information is most relevant. I expect this idea of leveraging agentic reasoning to continue to grow and become a key way that applications and agents search for data. That’s something we plan to bring into Redis as well, effectively introducing a new way to query your data.

Lastly, tell us a bit about the team at Redis. How would you describe your culture, and are you hiring? What do you look for in prospective team members joining the company? 

We are definitely hiring as a company, especially across the AI team. We have multiple teams on the AI front focused on search, on building new services like LangCache, and on what we’re doing around the context engine. We also have an AI research team. We’re hiring across all of these areas.

What we look for, beyond being technically strong, which is part of the culture at Redis, is someone who can adapt very quickly and has strong learning agility. That is especially critical in AI, where the stack and the technologies are evolving constantly and you need to keep pace with that change. Learning agility is a key requirement for anyone joining these teams.

Anything else you'd like our readers to know about the work you’re doing at Redis?

We’re really just getting started when it comes to building the context engine for the AI stack. We’re going to launch many more products and add new features across our existing offerings. The Redis you see today is already very different from what it was two years ago, and I’d say the Redis two years from now will be very different again, given the pace at which we’re evolving.

Conclusion

Stay up to date on the latest with Redis, follow them here.

Read our past few Deep Dives below:

• Arcade - A New Kind of Marketplace 🔍

• Strong Compute - Your High-Performance Training Infra Platform 🔋

• V7 Go - AI Agents for Complex Document Workflows 🎛

• Zencoder’s Repo Grokking - Your AI coding unlock 🔑

• Rime - Authentic AI Voices for Real-time Conversations at Scale 🗣️

If you would like us to ‘Deep Dive’ a founder, team or product launch, DM our chatbot here.

CV • Instagram • X • All Events