Recent months at Nextdata have been a whirlwind of building, hiring, and customer problem-solving. We’ve been so focused on our launch that we haven’t had time to respond to many of the questions friends and colleagues have asked us. We can’t get to everything, but this Q&A with our founder, Zhamak Dehghani, should answer many of them.

How are you doing?

I’m OK! In my career, I’ve never been this busy, nor have I ever been this excited about what I’m working on. I’m so thankful for my husband, who is doing an amazing job keeping me sane and caring for our daughter. 

How is hiring going?

It’s going incredibly well. We now have a strong leadership team in place, hyper-focused on our 2024 goals and committed to the company vision. Jonathon Morgan, our head of product, brings a wealth of experience as an ex-founder and product leader in the machine learning and enterprise data spaces. The tenacious Dr. Awdah Arraf is leading our operations, and Dr. Jörg Schad, an expert in distributed databases and graph ML, is leading our engineering. Sina Jahan, who has been in the trenches with me from the early days of data mesh, continues to shape the product and deliver value for our customers.

We have been inundated with applicants and inbound hiring requests, which is an honor but doesn’t make the process any easier. We’re looking for experienced, mission-driven team members who can flex and adapt to the shifting circumstances of an early-stage startup, and care about their craft. That’s hard to find. If you’re a rebel engineer who cares about decentralizing and democratizing access to data, we definitely want to hear from you! See our Careers page or contact us directly. 

How is the generative AI revolution impacting Nextdata?

Generative AI is shaping what’s happening with Nextdata in two big ways.

First, it’s increasing the urgency for solutions to manage decentralized data at scale. The promise of enterprise AI is real, but only if data currently dispersed across a dizzying array of locations can be made accessible to LLMs. You can’t simply gather every word spoken in a meeting or written in a document and dump it into a data lake. 

LLMs also require a much greater volume of data. Gathering the data required for training a model used to be a simple matter of running some queries and serializing query results, but with GenAI, we’re seeing the required data rates jump by ~6 orders of magnitude. 

The shape of data also becomes much more complex: training datasets; benchmarks and evals; preference optimization to fine-tune based on expert feedback; audits and guardrails for bias, safety, and other risks; and so on. Plus, with the popularity of Retrieval Augmented Generation (RAG) there are more immediate peer-to-peer needs for one department to fine-tune models or create embeddings at scale by consuming data products from other departments.

And we shouldn’t forget that LLMs themselves will generate a gusher of new data! Historically, the data available to train or augment AI/analytics has already been increasing by 1000X every 5 years. Generative AI is now dramatically increasing this growth in data. A decentralized approach to data/AI is the only path forward.

Enterprise data leaders are realizing that they need to embrace distributed, autonomous data if they have any hope of delivering on the AI use cases everyone is excited about. I’m excited to see that, and it’s already a big driver of our customer conversations.

nextdata OS is made for this scenario, with independent and decentralized computational data products running on different technology stacks and serving data and metadata together from a variety of storage types, serving LLM-friendly APIs natively.

The second impact of AI is within our product. We are finding ways to deploy generative AI and dramatically speed up the time to value of nextdata OS. AI really does make magical moments possible. I’ve been thrilled to see our team already finding incredible ways for us to leverage AI, and we’re only just getting started. 

What have you learned from your first wave of customers?

We have been lucky to work with some amazing customers, including some of the world’s most forward-thinking enterprise data leaders and teams. As we dig in, I continue to be disappointed by products that claim to implement the principles of data mesh but simply don’t.

I don’t blame customers for feeling frustrated by this gap in implementation. We’re working incredibly hard to close it. As I noted, AI will be a big enabler for us to move faster, hyperscale data mesh, and demonstrate the value of our approach. Getting a closeup view of the gaps left by current tooling has helped us refine our bootstrapping process, understand macro and micro-level use cases and sources of friction, and build real value into our product. 

We are encountering tension between getting results quickly and catalyzing meaningful behavior change, which takes time. Adopting data mesh asks people, teams, and organizations to work differently. Our clients have challenged us to find ways to kickstart the process and generate early results that help compel the larger behavioral shift. I’m proud to say that we’re rising to the challenge, and making effective use of AI in the process.  

What about that “Data mesh is dead” narrative?

Contrary to some loud voices in the industry (mostly analysts who are neither practitioners nor organizational decision-makers), Data mesh is alive and well. Whether we call it data mesh or not, the transition to decentralized,  domain-oriented data ownership, treating data as a product, and adopting data-mesh-native technology has begun, and it's going to be hard to stop.

As I meet with enterprise data leaders and learn about their challenges, it’s clear to me that data mesh is more important than ever. 

I can, however, understand where some of the criticism of data mesh comes from. I mentioned my frustration with superficial, marketing-driven products that claim to deliver on the principles of data mesh, then fail completely.

Some of these products are nothing more than money grabs created to cash in on hype. And, as should be expected, their failure has created a backlash. The frustration is deserved, and I feel it too. If anything about data mesh is dead, it’s customers’ tolerance for products that don’t deliver to their message.

But let’s not throw the baby out with the bathwater. As I’ve mentioned, the notion that we can simply dump all enterprise data in a data lake and LLMs will know what to do with it is totally unrealistic. I also believe the idea that we can simply “flow data through the pipelines” to get closer to compute is flawed. Modern data infrastructure built for structured data simply wasn’t built for the age of GenAI.

The context that will train effective enterprise LLMs largely lives in distributed, unstructured, hard-to-access data. Data mesh is the right approach to make that data accessible to everyone and everything that needs to use it, particularly AI and machine learning tools like LLMs.

What's the philosophy behind product development at Nextdata?

In a world where the incremental cost of adding each new feature trends lower and lower, category-leading product companies will be defined by what they don’t build as much as what they do. Our brilliant head of product, Jonathon Morgan, and I place a strong emphasis on discipline, focus, and simplicity. We have a structured filtering process where each potential feature must run a gauntlet of “so whats” before we consider it seriously. And everything must connect back to our vision and purpose. 

Our vision is to build a world where AI/ML and analytics are powered by decentralized, responsible, and equitable data ownership. Our purpose is to change the experience of creating, sharing, discovering, and using data forever, to be connected, fast, and fair.

The generalized rush to implement AI has led many companies to build features that impress people but don’t deliver real value. We refuse to fall victim to that compulsion. 

It’s inspiring to live in a moment where technology can feel magical again. And we definitely want magic in our product. But for us, the true magic happens when our customers and users see—finally—the promise of data mesh, realized.

If that’s exciting for you, please contact us. We need the right people on our team, and we need the right customer partners to help nextdata OS realize its tremendous potential.