I sat down with Rishi Sharma to learn more about the people behind Posh’s Natural Language Processing engine which powers one of FinTech’s most disruptive products: a chatbot for Credit Unions. Immediately, Rishi quickly clarifies: Posh isn’t just building chatbots; it’s fostering natural conversations. Rather than robotically spew answers, Posh’s chatbots seek to understand and delight users, remembering and reincorporating previous parts of a conversation while conveying human connection in addition to quick competence. Rishi is passionately invested in furthering Posh’s mission to make conversational AI more natural because human-like user experiences beget better technology at large. People and technology have never been more intertwined which is why I’m excited to speak with one of Posh’s most technically adept today.
When it comes to NLP technologies understanding context, Rishi speaks with authority. His method (detailed in this paper: https://www.aclweb.org/anthology/P18-2119.pdf) was used to measure the performance of state-of-the-art NLP Systems from research institutions and tech companies like Google. The most interesting part of this work? While a few were okay, most performed poorly, and none were better than humans. To better understand Rishi and Posh, I wanted to understand Rishi’s journey, and so we started where it all began.
“A band teacher”, Rishi says without hesitation. I don’t know what I expected, but that wasn’t it. Rishi immediately gushes about how cool his band teacher, “Mr. P,” was. As it turns out, this wasn’t a phase. Even today, music is a big part of Rishi’s life, and he is often playing his guitar and keyboard. When I asked him why it means so much to him, he took a reflective pause. Then he explained that music is a medium for connecting with other people. It’s a language totally different than the way we communicate with words. Maybe there was more that connected Rishi’s early desire to be a music teacher and his current mission of making language more natural than I at first thought.
As I describe my question, Rishi takes a beat to explain to me that “too many people say they aren’t Math people and let themselves be defined by that.” People are shaped by so many external factors, Rishi goes on to tell me, and his optimism shines as he takes the stance that anyone can be a Math person. Turns out this topic is personal to Rishi because he didn’t always consider himself a Math person. For his first two years of undergraduate study, Rishi was pursuing a major in History. History spoke to Rishi because of its inherent storytelling and the sense of exploration — there was always a new perspective to understand. His drive to understand multiple, sometimes innumerable, perspectives drove him to realize that he, too, was a Math person. As he grew frustrated by inherent bias in traditional languages like English. Rishi realized Mathematics was a neutral language that could communicate stories. It wasn’t just a set of procedures that “you’ll have to know someday,” but rather a language that could explain the world that came before him, the world around him now, and possibly the world to come.
It was through this question that I started to understand the power of the Posh vision. The chatbots that Posh are building have two points of value to their users, and coincidentally, they fall into both recurring themes of Rishi’s life: language and storytelling.
The internet has always been a collection of information. In the early days, the onus was on the user the navigate the computer’s natural language of directories and database structures. Over time, search became a dominant way to find what users wanted, but now finally, computers are meeting users where they are. Today’s chatbots help users find the solution they are looking for, but Rishi envisions a world where users describe the problem they are experiencing using their natural language and then bots interpret, ask followup questions, and solve users’ problems in a way a super-powered human might. This vision is where NLP elevates beyond a funnel conversion improvement or cost reducer into something profound. Children born today will ask “what do you mean you had to talk to people? How did a person possibly know all the answers?” and we’ll be left to respond: “well, they didn’t, and at times it was very frustrating.”
Rishi is obsessed that understanding the context of people is required to help them. Current algorithms most often take what someone most recently said or asked and analyze it through dozens, hundreds, or thousands of isolated dimensions. While that sounds complicated, it’s the opposite of how humans understand the world around them. Humans don’t think in ultra high-dimensioned, ephemeral moments. We think through stories — we analyze what few common things have led up to this moment. To make conversations more natural for humans, we need to build it to think with context and through stories.
Rishi hits me with some philosophy that underpins the company. In order to build the best products for people, we must understand people, and that means understanding all the things that make us great and the things that don’t. While the nature of the work keeps Rishi and the team looking outward at the cutting edge research that drives NLP, the culture of Posh keeps Rishi and the team looking inward and to each other to better understand people.
As we finish our conversation, I thank Rishi for sharing so much of his story with me, reflecting that this conversation was not what I expected. Rishi is obsessed with stories and how we communicate them. Taken in isolation, most of his past doesn’t seem like the path to being an NLP Engineer at one of Boston’s fastest-growing young companies, but after all, that’s his point. You don’t build stories from moments, but rather from finding the common thread between those moments.