Igor Jablokov is the CEO and Founder of Pryon. Named an “Industry Luminary” by Speech Technology Magazine, he previously founded industry pioneer Yap, the world’s first high-accuracy, fully-automated cloud platform for voice recognition. After its products were deployed by dozens of enterprises, the company became Amazon’s first AI-related acquisition. The firm’s inventions then served as the nucleus for follow-on products such as Alexa, Echo, and Fire TV. As a Program Director at IBM, Igor led the team that designed the precursor to Watson and developed the world’s first multimodal Web browser.
Igor was awarded Eisenhower and Truman National Security fellowships to explore and expand the role of entrepreneurship and venture capital in addressing geopolitical concerns. As an innovator in human language technologies, he believes in fostering career and educational opportunities for others entering STEM fields. As such, he serves as a mentor in the TechStars’ Alexa Accelerator, was a Blackstone NC Entrepreneur-In-Residence (EIR), and founded a chapter of the Global Shapers, a program of the World Economic Forum.
Igor holds a B.S. in Computer Engineering from The Pennsylvania State University, where he was named an Outstanding Engineering Alumnus, and an MBA from The University of North Carolina.
Your journey in AI started with the first cloud-based speech recognition engine at Yap, later acquired by Amazon. How did that experience shape your vision for AI and influence your current work at Pryon?
I’ll start a bit earlier in my career as Yap wasn’t our first rodeo in dealing with natural language interactions.
My first foray into natural language interactions started at IBM, where I started as an intern in the early 90s and eventually became Program Director of Multimodal Research. There I had a team that discovered what you could consider a baby Watson. It was far ahead of its time, but IBM never greenlit it. Eventually I became frustrated with the decision and departed.
Around that time (2006), I recruited top engineers and scientists from Broadcom, IBM, Intel, Microsoft, Nuance, NVIDIA and more to start the first AI cloud company, Yap. We quickly acquired dozens of enterprise and carrier customers, including Sprint and Microsoft, and almost 50,000,000 users on the platform.
Since we had former iPod engineers on the team, we were able to back-channel into Apple within a year of founding the company. They brought us in to prototype a version of Siri—this was before the iPhone was released. Half a decade later, we were secretly acquired by Amazon to develop Alexa for them.
Can you elaborate on the concept of “knowledge friction” that Pryon aims to solve and why it’s crucial for modern enterprises?
Knowledge friction comes from the fact that, historically, organizations haven’t had one unified instantiation of knowledge. While we’ve had such repositories in our college campuses and civic communities in the form of libraries, there has been no unification of data and knowledge on the enterprise side due to a myriad of vendors they used.
As a result, everyone across virtually every organization feels friction when looking for the information they need to perform their jobs and workflows. This is where we saw the opportunity for Pryon. We thought that there was an opportunity for a new layer above the enterprise software stack that, by using natural language prompts, could traverse systems of records and retrieve various object types—text, images, videos, structured and unstructured data—and pull everything together in a sub-second response time.
That was the birth of Pryon, the world’s first AI-enhanced knowledge cloud.
Pryon’s platform integrates advanced AI technologies like computer vision and large language models. Can you explain how these components work together to enhance knowledge management?
Pryon developed an AIP, an artificial intelligence platform, that transforms content from its fundamental static units into interactive knowledge. It achieves this by integrating an ingestion pipeline, a retrieval pipeline, and a generative pipeline into a single experience. The platform taps into your existing systems of record, which can include a variety of content types such as Confluence, Documentum, SAP, ServiceNow, Salesforce, SharePoint, and many more. This content can be in the form of audio, video, images, text, PowerPoints, PDFs, Word files, and web pages.
The AIP transforms these objects into a knowledge cloud, which can then publish and subscribe to any interactive or sensory experiences you may need. Whether people need to interact with this knowledge or there are machine-to-machine transactions requiring the union of all this disparate knowledge, the platform ensures consistency and accessibility. Essentially, it performs ETL (Extract, Transform, Load) on the left side, powering experiences via APIs on the right side.
What are some of the key challenges Pryon faces in developing AI solutions for enterprise use, and how are you addressing them?
Because we are vertically integrated, we receive top marks in accuracy, scale, security, and speed. One of the issues with deconstructed approaches, where you need several different vendors and bolt them together to achieve the same workflow we do, is that you end up with something less performant. You can’t match models, and you don’t have security signaling flowing through as well.
It’s like iPhones: there’s a reason Apple builds their own chip, device, operating system, and applications. By doing so, they achieve the highest level of performance with the lowest energy use. In contrast, other vendors who integrate from several different sources tend to be a generation or two behind them at all times.
How does Pryon ensure the accuracy, scalability, security, and speed of its AI solutions, particularly in large-scale enterprise environments?
Supported by a robust Retrieval-Augmented Generation (RAG) framework, Pryon was designed to meet the rigorous demands of businesses. Using best-in-class information retrieval technology, Pryon securely delivers accurate, timely answers — empowering businesses to overcome knowledge friction.
- Accuracy: Pryon excels in accuracy by precisely ingesting and understanding content stored in various formats, including text, images, audio, and video. Using advanced custom-developed technologies, Pryon retrieves mission-critical knowledge with over 90% accuracy and delivers answers with clear attribution to source documents. This ensures that the information provided is both reliable and verifiable.
- Enterprise Scale: Pryon is built to handle large-scale enterprise environments. It scales to millions of pages of content and supports thousands of concurrent users. Pryon also includes out-of-the-box connectors to major platforms like SharePoint, ServiceNow, Amazon S3, Box, and more, making it easy to integrate into existing workflows and systems.
- Security: Security is a top priority for Pryon. It protects against data leaks through document-level access controls and ensures that AI models are not trained on customer data. Furthermore, Pryon can be implemented in on-premises environments, offering additional layers of security and control for sensitive information.
- Speed: Pryon offers rapid deployment, with implementation possible in as little as two weeks. The platform features a no-code interface for updating content, allowing for quick and easy modifications. Additionally, Pryon provides the flexibility to choose a public, custom, or Pryon-developed large language model (LLM), making the implementation process seamless and highly customizable.
This is why academic institutions, Fortune 500 companies, government agencies, and NGOs in critical sectors like defense, energy, financial services, and semiconductors leverage us.
Pryon emphasizes Responsible AI with initiatives like respecting authorship and ethical sourcing of training data. How do you implement these principles in your day-to-day operations?
Our clients and partners control what goes into their instance of Pryon. This includes public information from trusted academic institutions and government agencies, published information they’ve properly licensed for their organizations, proprietary information that forms the core IP of their business, and personal content for individual use. Pryon synthesizes these four source types into a unified knowledge cloud, completely under the control of the sponsoring organization. This ability to securely manage diverse content types is why we’re trusted in robust environments, including critical infrastructure.
With Pryon recently securing $100 million in Series B funding, what are your top priorities for the company’s growth and innovation in the coming years?
Post-Series B, we’re in early growth territory. One part of this phase is industrializing the product market fit we’ve established to support the cloud environments and server types our clients and partners are likely to encounter.
The first focal area is ensuring our product can handle these demands while also offering them modular access to our capabilities to support their workflows.
The second major area is developing scaling partners who can build practices around our work with our tooling and manage the necessary change as organizations transform to support the new era of digital intelligence. The third focus is continued R&D to stay ahead of the curve and define the state of the art in this space.
As someone who has been at the forefront of AI innovation, how do you view the current state of AI regulation, and what role do you believe Pryon can play in shaping these discussions?
I think we all wonder how the world would have turned out if we had been able to regulate some technologies closer to their infancy, like social media, an example. We didn’t realize how much it would affect our communities. Different nation-states have different perspectives on regulation. The Europeans have a somewhat constrained perspective that matches their values with the EU AI Act.
On the flip side, some environments are completely unconstrained. In the US, we’re looking for a balance between allowing innovation to thrive, especially in commercial activities, and safeguarding sensitive use cases to avoid biases and other risks, such as in approving loan applications.
Most regulation tends to target the most sensitive use cases, particularly in consumer applications and public sector or government uses. Personally, that’s why I’m on the board of With Honor, a bipartisan coalition of veterans, policymakers, and lawmakers. We have seen convergence, regardless of political beliefs, on concerns about the introduction of AI technologies into all aspects of our lives. Part of our role is to influence the evolution of regulation, providing feedback to find the right balance we all wanted for other technology areas.
What advice would you give to other AI entrepreneurs looking to build impactful and responsible AI solutions?
Right now, it’s going to be both a wild west and a fantastical environment for developing new forms of AI applications. If you don’t have extensive experience in AI—say, 10, 20, or 30 years—I wouldn’t recommend developing an AI platform from scratch. Instead, find an application area where the technology intersects with your subject matter expertise.
Whether you’re an artist, attorney, engineer, lineman, physician, or in another field, leveraging your expertise will give you a unique voice, perspective, and product in the marketplace. This approach is likely to be the best use of your time, energy, and experience, rather than creating another “me too” product.
Thank you for the great interview, readers who wish to learn more should visit Pryon.
