Birago Jones is the CEO and Co-Founder of Pienso, a no-code/low-code platform for enterprises to train and deploy AI models without the need for advanced data science or programming skills. Today, Birago’s customers include the US government and Sky, the largest broadcaster in the UK. Pienso is based on Birago’s research from the Massachusetts Institute of Technology (MIT), where he and his co-founder Karthik Dinakar served as research assistants in the MIT Media Lab. He is a distinguished authority in the intersection of artificial intelligence (AI) and human-computer interaction (HCI), and an advocate for responsible AI.
Pienso‘s interactive learning interface is designed to enable users to harness AI to its fullest potential without any coding. The platform guides users through the process of training and deploying large language models (LLMs) that are imprinted with their expertise and fine-tuned to answer their specific questions.
What initially attracted you to pursue your studies in AI, HCI (Human Computer Interaction) and user experience?
I had already been developing personal projects focused on creating accessibility tools and applications for the blind, such as a haptic digital braille reader using a smartphone and an indoor wayfinding system (digital cane). I believed AI could enhance and support these efforts.
Pienso was initially conceived during your time at MIT, how did the concept of training machine learning models to be accessible to non-technical users originate?
My co-founder Karthik and I met in grad school while we were both conducting research in the MIT Media Lab. We had teamed up for a class project to build a tool that would help social media platforms moderate and flag bullying content. The tool was gaining lots of traction, and we were even invited to the White House to give a demonstration of the technology during a cyberbullying summit.
There was just one problem: while the model itself worked the way it was supposed to, it wasn’t trained on the right data, so it wasn’t able to identify harmful content that used teenage slang. Karthik and I were working together to figure out a solution, and we later realized that we could fix this issue if we found a way for teenagers to directly train the model data.
This was the “Aha” moment that would later inspire Pienso: subject-matter experts, not AI engineers like us, should be able to more easily provide input on model training data. We ended up developing point-and-click tools that allow non-experts to train large amounts of data at scale. We then took this technology to local Cambridge, Massachusetts schools and elicited the help of local teenagers to train their algorithms, which allowed us to capture more nuance in the algorithms than previously possible. With this technology, we went to work with organizations like MTV and Brigham and Women’s Hospital.
Could you share the genesis story of how Pienso was then spun out of MIT into its own company?
We always knew that this technology could provide value beyond the use case we built, but it wasn’t until 2016 that we finally made the jump to commercialize it, when Karthik completed his PhD. By that time, deep learning was exploding in popularity, but it was mainly AI engineers who were putting it to use because nobody else had the expertise to train and serve these models.
What are the key innovations and algorithms that enable Pienso’s no-code interface for building AI models? How does Pienso ensure that domain experts, without technical background, can effectively train AI models?
Pienso eliminates the barriers of “MLOps” — data cleaning, data labeling, model training and deployment. Our platform uses a semi-supervised machine learning approach, which allows users to start with unlabeled training data and then use human expertise to annotate large volumes of text data rapidly and accurately without having to write any code. This process trains deep learning models which are capable of accurately classifying and generating new text.
How does Pienso offer customization in AI model development to cater to the specific needs of different organizations?
We are strong believers that no one model can solve every problem for every company. We need to be able to build and train custom models if we want AI to understand the nuances of each specific company and use case. That’s why Pienso makes it possible to train models directly on an organization’s own data. This alleviates the privacy concerns of using foundational models, and can also deliver more accurate insights.
Pienso also integrates with existing enterprise systems through APIs, allowing inference results to be delivered in different formats. Pienso can also operate without relying on third-party services or APIs, meaning that data never needs to be transmitted outside of a secure environment. It can be deployed on major cloud providers as well as on-premise, making it an ideal fit for industries that require strong security and compliance practices, such as government agencies or finance.
How do you see the platform evolving in the next few years?
In the next few years, Pienso will continue to evolve by focusing on even greater scalability and efficiency. As the demand for high-volume text analytics grows, we’ll enhance our ability to handle larger datasets with faster inference times and more complex analysis. We’re also committed to reducing the costs associated with scaling large language models to ensure enterprises get value without compromising on speed or accuracy.
We’ll also push further into democratizing AI. Pienso is already a no-code/low-code platform, but we envision expanding the accessibility of our tools even more. We’ll continuously refine our interface so that a broader range of users, from business analysts to technical teams, can continue to train, tune, and deploy models without needing deep technical expertise.
As we work with more customers across diverse industries, Pienso will adapt to offer more tailored solutions. Whether it’s finance, healthcare, or government, our platform will evolve to incorporate industry-specific templates and modules to help users fine-tune their models more effectively for their specific use cases.
Pienso will become even more integrated within the broader AI ecosystem, seamlessly working alongside the solutions / tools from the major cloud providers and on-premise solutions. We’ll focus on building stronger integrations with other data platforms and tools, enabling a more cohesive AI workflow that fits into existing enterprise tech stacks.
Thank you for the great interview, readers who wish to learn more should visit Pienso.
Credit: Source link
