Sarah Nagy is the founder and CEO of Seek AI, a platform that enables business end-users to ask Seek the exact same questions that they currently ask the data team, right in Slack, Teams and email. No “finessing” of how they write their question, and no learning a new platform.
You initially started as a researcher with data from the Hubble Space Telescope. What were you working on?
I was doing research at UCLA and Caltech, looking at some of the most distant galaxies that were able to be observed with a telescope, and was working on analyzing some of their properties such as their mass and size. The purpose of this research was to help us understand the difference between very distant galaxies versus galaxies that are closer to our own, and develop models for how these galaxies form over time.
You then worked as a data scientist at various startups. What were some of the more interesting projects?
One project that stands out involved using natural language processing (NLP) to classify unstructured text relating to retail items. For example, taking raw text (e.g. “air jordans green”) and labeling as the estimated brand (“Nike”). I had a colleague who specialized in NLP that was busy with a different project, so I actually wasn’t originally supposed to work on this one. It ended up being handed to me since they were busy. I didn’t even know anything about NLP at the time, so I went through some free courses from Stanford and Fast.ai to ramp up my knowledge. I really enjoyed learning about NLP and started to understand why it is so important, and why artificial intelligence (AI) being able to understand language is a big step towards so-called “general AI.” This experience definitely primed me to be quick to understand the importance of GPT-3 when it first came out.
Could you share the genesis story behind Seek AI?
When OpenAI’s GPT-3 model came out, I immediately recognized what an incredible advancement it was and got particularly excited about applications involving GPT-3 writing code. After all, I was writing code all day as a data scientist, and to see AI doing this – and generating the code perfectly – was jaw-dropping. I’d compare my reaction to GPT-3 to first learning about VR back in 2013, which was another jaw-dropping experience for me. I ended up deciding that I needed to form a startup to make a bet on this technology. I didn’t know exactly what I was going to build, but I had a gut feeling that if I learned more about these models, something valuable would fall into place.
Once I had really learned about the models, that’s when I realized I could solve a pain point I encountered everywhere I had worked as a quant or as a data scientist. The pain point in question was business people not having the right tools to answer their own data questions. As a data scientist, I would frequently work on problems that required a lot of focus, but I was often interrupted by colleagues on the business side who had questions about the data, forcing me to stop what I was doing. The process seemed archaic and inefficient. I realized that if I focused on this new technology solving the problem, it would be a category-defining solution to this very important and ubiquitous problem.
Seek AI uses generative AI. Could you explain to our readers what this is?
“Generative AI” is a very hyped buzzword, but unlike other buzzwords, I don’t believe the hype is unwarranted. The term refers to large machine learning models with hundreds of billions of parameters, such as Open AI’s DALL-E and GPT-3. The innovation of these models is that they can understand natural language and generate text, images, code, and more. If you ever play around with DALL-E or Stable Diffusion, for example, you will quickly understand why these models are so hyped; they have an incredibly human-like ability to understand natural language commands and can generate art that rivals the best human artists.
Code generation is one of the most niche, but most important, applications of generative AI. Data is getting bigger and more complex, and therefore harder to manually analyze and organize by humans. Yet, there is so much information encoded in this data. This information is not just powerful for organizations, it can also lead to incredible scientific breakthroughs on the academic side. Building AI to extract value from data will unlock incredible value in the form of useful information.
Seek AI is building an interface that enables users to interact with data using natural language. Knowledge workers can access Seek AI’s natural language interface by means of email, Slack, text, and a range of customer relationship management (CRM) systems.
What other types of machine learning are used at Seek AI?
While generative AI is a piece of our machine learning architecture, our architecture also includes several forks of open-source deep learning models. Transformer models (of which “generative AI” is a variant) comprise many (but not all) of the models that Seek uses.
Why is it so important for non-technical users to be able to rapidly access data?
What good is data if it’s not generating an ROI, and how can a business get this ROI if business-facing users can’t even access it? This is why it is absolutely essential to give access to as many people as possible, without compromising accuracy.
When I was a data scientist, sometimes I would get requests from the CEO to analyze some data to help with our company’s product or go-to-market strategy. These projects could take weeks or longer. As a CEO now, I definitely understand the importance of those projects at a deeper level than I did when I was on the data side. I often find myself wishing that I could simply get the data at my fingertips so I can make my decisions faster. This is an example of what we are solving at Seek.
How does Seek AI make this data so easy to retrieve?
Something that is interesting to think about is that data can really only be analyzed with code. It’s true that there are platforms that are abstractions over this code (e.g. data dashboards), but under the hood, there is code manually written by data analysts which enables the data to be presented to the business end users.
Most knowledge workers don’t know how to code, don’t want to code, or simply can’t even get access to the data even if they do want to write code to analyze it. Therefore, when they need data, they either need to locate it in a dashboard or ask the data team if they can’t find it. The bigger that datasets get, the more this will happen.
Data teams therefore need to be “translators” of natural language questions directed to them, and the data itself, which they query using code. Removing this “translator” intermediary is the heart of what Seek is doing.
How do enterprises ensure that the data that they use is accurate?
Managing the tradeoff between data accuracy and accessibility is a huge challenge. As I stated in a recent interview, on one hand, accessibility allows less technical folks to start interacting with the knowledge wellspring that is a company’s data. On the other hand, what good is a wellspring of polluted water (i.e. bad data)?
The best data teams are those that manage this tradeoff in the most optimal way possible, and a big part of that is carefully calibrating and vetting any tools that non-technical users can interact with.
What are some examples of use cases for the Seek AI platform?
We are already delivering value to customers and design partners in the B2B SaaS, Fintech, Consumer Product Goods (CPG), and B2C e-commerce vertical markets.
Battlefin, for example, is the leading marketplace of alternative financial datasets. They believe that giving fast, high-quality answers to their own customers’ questions is the difference between winning and losing over their competitors. The company’s CEO, Tim Harrington, noted, “Seek AI played a critical role in our company’s 2023 strategy because of the edge that it gives us in accessing and analyzing our 2,400+ datasets in response to customer questions. I’d estimate that our ROI on Seek AI is about 10x based on what we would have spent to achieve this level of efficiency without the platform.”
Is there anything else that you would like to share about Seek AI?
This might be the right place for a shameless plug. Seek is currently offering free trials of our platform, which can be accessed on seek.ai. We’re excited to be a pioneer in bringing generative AI to data teams, and I am looking forward to going on this journey with our customers.
Thank you for the great interview, readers who wish to learn more should visit Seek AI.
Christian Noske, Partner at NGP Capital – Interview Series
Antoine Tardif is a Futurist who is passionate about the future of AI and robotics. He is the CEO of BlockVentures.com, and has invested in over 50 AI & blockchain projects. He is the Co-Founder of Securities.io a news website focusing on digital assets, digital securities and investing. He is a founding partner of unite.AI & a member of the Forbes Technology Council.
Christian Noske, Partner at NGP Capital – Interview Series
Avinash Misra, CEO & Co-Founder of Skan.AI – Interview Series
Krish Ramineni, Co-Founder & CEO of Fireflies.ai – Interview Series
Liran Hason, Co-Founder & CEO of Aporia – Interview Series
Roshanak Houmanfar, VP of Machine Learning Products at Integrate.ai – Interview Series
Ronny Shalev, PhD, CEO & Founder of Dyad Medical Inc – Interview Series
Advertiser Disclosure: Unite.AI is committed to rigorous editorial standards to provide our readers with accurate information and news. We may receive compensation when you click on links to products we reviewed.
Copyright © 2021 Unite.AI