Featureform: Streamlining How Data And Model Features Are Built In Machine Learning Organizations

By Amit Chowdhry ● Mar 14, 2024

Featureform is a company that turns features into a first-class component of the machine learning process. Pulse 2.0 interviewed Featureform CEO Simba Khadder to learn more about the company.

Simba Khadder’s Background

Khadder started his career at Google where he was a Software Engineer working on Cloud Datastore and Google Wide Profiling. And Khadder said:

“I left to co-found Triton, a media personalization and analytics platform, which handled over 100 million users.”

Featureform Co-Founders Simba Khadder and Shabnam Mokhtarani
(Featureform Co-Founders Simba Khadder and Shabnam Mokhtarani)

Formation Of Featureform

How did the idea for the company come together? Khadder shared:

“At my previous startup, we managed a variety of models, including recommender systems, for over a hundred million users. To enhance the productivity and collaboration of our Machine Learning (ML) teams and ensure the reliability and compliance of our models, we developed our own internal MLOps platform.”

“We discovered that a significant challenge in our process was feature engineering, which involves transforming raw data into useful inputs for our models. In practice, most of our model’s performance gains and most of our time came from working on the data. To make this process fast, collaborative, and reliable, we created an early version of what would become Featureform. It served as a dedicated data platform for our ML teams.”

Core Products

What are the company’s core products and features? Khadder explained:

“Featureform is the creator of the virtual feature store. Our mission is to streamline how data and model features are built and maintained in machine learning orgs. We make it easier for data scientists to deploy their features, iterate on them, and monitor them in production.”

“Our python framework and feature store organizes experimentation and fosters collaboration. It does away with copy and pasting between scattered notebooks with names like “Untitled18.ipynb”, unifies feature pipelines between experimentation and production, deduplicates repeated features across teams, and eliminates ambiguously named tables like “featuretablev5.” While we pride ourselves on our open-core model, we also offer a robust enterprise solution with governance, streaming, and more.”

Favorite Memory

What has been Khadder’s favorite memory working for the company so far? Khadder reflected:

“I vividly recall a moment during a call with one of our first customers when they shared their Featureform dashboard with us. The instance was full of activity, and their excitement was palpable. They were overjoyed by how it had revolutionized their team’s processes for the better. There’s a kind of magic in seeing the product you’ve dedicated so much effort to — your blood, sweat, and tears — create tangible value for a customer. It validates all the hard work, and it’s what energizes me to wake up and get going every morning!”

Challenges Faced

What bottlenecks has Khadder faced in his sector of work recently? Khadder acknowledged:

“It’s not so much a bottleneck as an evolution of the market. There has been a significant but exciting shift. In the initial stages of MLOps, there was considerable hype, which is typical with any new transformative tech. This enthusiasm was justified, though perhaps a bit premature, considering the MLOps market’s maturity at the time. Now, as the market has matured, we find ourselves in a stronger position, moving beyond the hype to focus squarely on delivering tangible value to our customers.”

“Interestingly, at the same time, AI and LLMs are injecting new excitement into a market that slightly overlaps with ours. This presents a fascinating challenge, but our strategy remains unchanged. We are steadfastly committed to grounding our work in real value creation. This approach has successfully guided us through the previous hype cycle, and I’m confident it will steer us through any future ones.”

“We didn’t just ride the MLOps wave and we won’t just ride the LLMOps wave; we’re shaping the future of these markets, ensuring that our customers always have the best tools and solutions to succeed in this dynamic landscape.”

Evolution Of Featureform’s Technology

How has the company’s technology evolved since launching? Khadder noted:

“Featureform’s goals have not changed since launching; they are: improving feature deployment, enhancing team collaboration, organizing feature experimentation, increasing reliability in production, and ensuring compliance in data handling. Our initial open-source offering was a strong foundation, and we’ve built upon it extensively since then. We release new updates monthly, continually expanding our capabilities.”

“This includes deeper integrations into platforms like Databricks and Snowflake for large-scale feature engineering in enterprise environments. We’ve also refined our versioning, lineage, and dashboard features for better collaboration and experimentation. Additionally, new monitoring features help in proactively maintaining model performance. On the compliance side, our enhanced integrations with identity providers like Okta and data catalogs ensure a smooth fit into any data governance framework. From an early open-source product, Featureform has evolved into a comprehensive, enterprise-grade platform.”

Significant Milestones

What have been some of the company’s most significant milestones? Khadder cited:

“One of our company’s major achievements has been the launch of our Enterprise product. We consistently encountered two primary needs from our customers: governance and stream processing. Addressing these needs required a lot of engineering effort. For streaming, we need to handle all the intricacies around backfill, scaling, and point-in-time correctness.”

“For governance, we needed to create flexible solutions that could integrate seamlessly into our clients’ existing governance and compliance tools. The release of our Enterprise product is a significant step for both the Featureform company and our open-source project, paving the way for long-term success. This milestone was made possible thanks to the dedicated hard work of our product team!”

Funding

After asking Khadder about the company’s funding information, he revealed:

“This last round of funding takes our total funding up to about $8m. We’re excited to have a new partner in GreatPoint Ventures and to have the continued support and follow-on funding from Zetta Venture Partners.”

Total Addressable Market

What total addressable market (TAM) size is the company pursuing? Khadder assessed:

“Our long-term vision is to become the Hashicorp of MLOps. The market is huge and it becomes clear as we look across different industries. Take financial institutions, for example. They use ML for everything from fraud detection to chatbots. The same goes for insurance companies, and when you look at commerce companies, the use extends even further into supply chain management and warehouse operations.”

“ML is becoming ubiquitous, and with the emergence of LLMs and generative AI, we’re only going to see this trend accelerate. At Featureform, we’re targeting a future where every Fortune 500 company leverages our platform to enhance the productivity and collaboration of their ML teams and to ensure their models are both reliable and compliant.”

“So, when we talk about our TAM, we’re looking at a landscape where nearly every major company across diverse sectors could be a potential customer. Our TAM is essentially as broad as the application of machine learning in the business world, which, as we see it, is an ever-growing space.”

Differentiation From The Competition

What differentiates the company from its competition? Khadder affirmed:

“What sets our company apart in the feature store market is our unique approach to feature store architecture. Unlike traditional feature stores, that we call ‘literal’ feature stores, we don’t just see a feature store as a storage layer for a feature table. Instead, we focus on storing the logic in creating the feature, and orchestrate its creation as well as monitor it in production. This perspective allows us to offer features like lineage, versioning, and improved collaboration for feature engineering.”

“Another class of feature stores, which we refer to as physical feature stores or feature platforms, require teams to shift their data onto a new platform and use a specific transformation engine. At Featureform, we adopt a virtual architecture. This approach makes us infrastructure-agnostic, giving teams the flexibility to select the best data infrastructure for their needs, while we provide an overlying feature platform.”

“Our strategy enables us to deliver comprehensive value throughout the entire feature lifecycle – from the initial idea to production. Importantly, we achieve this while keeping adoption costs low and maintaining being infrastructure-agnostic.”

Future Company Goals

What are some of the company’s future company goals? Khadder concluded:

“Our long-term ambition is to be recognized as the Hashicorp of MLOps. Our company’s name, Featureform, reflects our admiration and design inspiration from products like Terraform. This latest round of funding is an opportunity to amplify our investment in our product, ensuring that we continue delivering solutions that our customers love and rely on. But it’s not just about the product; we also see an essential role for ourselves in educating the market.”

“The MLOps sector is rapidly evolving, with technological advancements, market shifts, and varying levels of hype. This can lead to confusion and fragmentation, and we want to be a beacon of clarity and guidance. Ultimately, our goal is to support and accelerate more and more customer’s ML and AI initiatives.”