Under the Hood: Parmonic's AI Engine

Years of research and thousands of test videos have helped us develop a sophisticated, purpose-built AI engine to parse informational videos.

While the AI is very complicated, in this post we give you a high-level, understandable overview of what's happening under the hood of Parmonic's AI.‍

When people get on a video - they "show and tell". Videos are made up of two main ingredients - visuals and audio.

While humans can hear and see, computers cannot. So the first task is to convert videos into a format that computers can understand. To accomplish this our technology first extracts these two basic signals from every video file, and converts these signals into bits and bytes that our algorithms can act on.

A video is a large collection of static images - frames. We extract each frame and parse it. The spoken audio is transcribed into text. 

Next, the algorithms need to make sense of the content and understand it. This is one the hardest problems in AI. We've been working on it for more than 4 years now and have developed some of the best algorithms on the planet.

We call the algorithmic approach that we have developed as the Expert Ensemble Method. As the name implies, this approach uses several "experts" to find the key moments in a video. Each "expert" is a different model. We use several different AI models, as well as heuristics-based models. 

Our AI analyzes each frame to find the most interesting key moments. It then ranks the key moments and suggests those with the highest probability of being superior key moments.

UPDATE: In fall 2023, we received a patent from the USPTO which is a seal of approval of our novel and innovative approach to finding key moments. If you are a Parmonic customer or evaluating Parmonic, rest assured that you are in good hands.

Sidenote: one of the things that is unique to Parmonic is the experiences and heuristics that some of our team has from a B2B webinar and marketing content perspective. AI without domain knowledge is limited in what it can do. By combining our domain knowledge with AI expertise we've been able to solve some hard problems creatively.‍

The following animation shows a glimpse of how our models work in unison. There are a number of "experts" in our approach. One AI expert is observing the duration of each visual scene (or slide in case of webinars), while another is reading what's on the screen and using OCR. Yet another might be doing complex topic and keyword analysis. AI models are also paying attention to the number of speakers, Q&A pairs and more. 

The beauty of our Expert Ensemble approach is that we can add more experts over time and continuously refine and improve the outcomes. We have filed several patents to protect our research so that our customers can use a robust, enterprise-quality product.‍

‍

Under the hood, our algorithms are performing extremely complicated (and novel) NLP analysis, Deep Learning, Neuro-Symbolic AI and expert modeling.

But, we understand that while our AI engine is very sophisticated, what matters most to our customers is not the intricacies of the AI but the results and the impact that they can have at their companies. Embracing that philosophy we don't harp on buzzwords and try hard to distill the complexities to make our product easy to use.

A very important aspect of Parmonic is the ability to easily choose different moments from a video even if the AI didn't recommend those. Our Magic Highlighter makes this possible so users can highlight portions of the transcript and our engine builds videos based on that. This control ensures that users have the ability to fully own the outputs that get published. Our AI is your sidekick - you are the boss!

The most beautiful concepts in nature and technology are simple and complicated at the same time. If you're a marketer, join us in this journey and see how you can use AI to become an inspirational marketer who can convert long videos into moments that matter.‍