Engineering Wisdom

We land the member on a single piece of content, much like Tiktok does, in a “For You” tab that they can swipe through. We are always trying to show the right piece of content to the right community member. We believe our swipe-style mechanic is superior to Clubhouse’s feed of content, as our algorithms are able to learn more efficiently and accurately based on the explicit signals from each member.

We train our neural network to look for one thing: a conversation that matters to the listener. Wisdom strives to be more than yet another diversionary app. TikTok provides short clips meant to amuse and help pass the time. And they’re quite good at it. While employing similar algorithms and strategies to TikTok, Wisdom seeks a loftier goal: making them better humans.

We do this by providing the Wisdom audience with thought-provoking, wise content that is relevant to them. Most apps today train their algorithms with the goal of increasing the amount of time a “user” spends within the app. A byproduct of that goal is the algorithm suggests content that is interesting to the “user.”

At Wisdom, we think that relationship is backward. Our goal is to provide our members with conversations that matter, content of substance that can deepen their knowledge and improve their life. We take umbrage at the word “user” and what it implies. TikTok has users. Facebook has users. So does heroin. It’s not acceptable to us to create addictions to content that has worse than no merit. Tech companies and their employees need to ask themselves if the end goal of their product makes people’s lives better or worse.

In contrast, Wisdom is not after passive amusement and wasting incremental seconds of our audience’s valuable time. We want them to be engaged by our mentors and wise people. We want them to be better for the time spent listening. We deeply believe long-form, wise content is the antidote for our distraction-oriented technology-laden culture.

Wisdom provides both live conversations and previously recorded talks. The previously recorded talks allow us to provide a “Best Of” section to showcase the best content to each user. Both live and recorded conversations are also tagged with keywords and titled to help improve our algorithmic recommendations.

On the technical side of things, we employ both engineered and machine-learning algorithms. The system gathers metrics to continually compare the results of each algorithm in order to make decisions and adjustments. New algorithms can be added to the mix on the fly. The new algorithm is then measured against the other algorithms to determine its efficacy. Here, at the very beginning of Wisdom’s history, the complexity of the algorithm may not seem to be so powerful, but just wait! As more and more talks are created, the algorithm will get only better at surfacing the promise of Wisdom: conversations that matter.

Today we leverage a few different algorithms. We have a custom ML algorithm that powers the surfacing of the live talks. This is incredibly important to get right. Landing the right member on the right live conversation gives them the ability to speak directly with the expert as a guest. Next, we have a ML-based algorithm using AWS Personalize. This powers the “Best Of” section. Additionally for the “Best Of” section, we have algorithms that are customized combinations of SVD (singular value decomposition), NMF (Non-negative Matrix Factorization), and DNN (deep neural network).

The ML algorithms are based on features of the content itself and of the activity we gather from the platform. Content features are not only title, tags, and member-reported information, but we also process the talk itself to extract the topics being discussed and the sentiment. We do this using audio and text transformer models such as BART, BERT, and RoBERTa in real-time. For the activity feature, we look at our members’ actions such as joining a talk, leaving a talk (and the time duration between those two events), whether they chose to follow the host, or to unfollow them.

We care deeply about improving our tech and getting better and better at applying the state-of-the-art in machine learning, not because we are interested in technology per se, but because we believe in the promise of technology to make us more human, to link poetry with processors, to create an automated way to not only tap into the wisdom of humanity but to shape that wisdom.

Related posts:

Popularly termed Industry 4.0, the AI revolution is an age-long concept existing for more than a decade now. Its use case is as expansive and transformative as ever. This is why Big tech Companies…

Working hard versus Working smart is always debatable. However, I have a different opinion over considering them as two workstyles. There must be an amalgamation of the two. Hard work is necessary to…

I often feel our brains have clocks of their own. Time is after all, just the fourth dimension. Whether it really exists, is a question for another time. In my so many years of existence, I've come…