Kensho NLP hackathon: Exploring new challenges with natural language processing
Kensho's first-ever NLP hackathon brought engineers, designers, and PMs together to build experimental AI solutions.
Kensho’s NLP team recently organized the company’s first ever NLP-specific hackathon that brought together engineers, designers and product managers to explore and develop experimental products using NLP techniques. The hackathon was a great team-building event for connecting, collaborating, showcasing skills and creating new solutions.
“I’m really impressed with what our teams were able to deliver in only a few short days,” said Bhavesh Dayalji, CEO of Kensho Technologies. “I’m encouraged by how we continue to experiment and push the boundaries of what is possible with our technology. This spirit of innovation is what makes it so exciting to be working where we are right now.”
Over 20 Kenshins participated across four teams, with each team consisting of two machine learning engineers, one software engineer and one product manager or designer. Participants worked intensively over a 72-hour period to create functional prototypes, concluding with a company-wide presentation on April 20.
“The thoughtfulness and enthusiasm with which the teams approached the work translated to results that were remarkable for such a short period of time,” said Zak Brown, Kensho’s NLP Team Lead. “Everyone is excited about the many real-life applications of these projects.”
Read on to learn more about the various projects.
Document question & answering
Lupe, Arijit, Rik and Matt teamed up to develop a question-answering prototype related to long-form financial documents with rich structure. Their model allows users to ask questions of financial documents such as earnings calls to find and extract important data points. As a result, users can quickly find the relevant information they need from long financial documents and make it readily available for downstream consumption.
A question-answering workflow showing questions generated from a document
Financial News Summarization
The financial news summarization team, consisting of Guillaume, Yash, Qibo and Victoria, focused on building a natural language processing text summarization tool for financial news. Their tool allows users to break down longer documents into key salient points. Through an intuitive interface, a user inputs their body of text and, with a single click, receives a model-generated summary, as well as the option to provide feedback.
The Kensho Summarization tool, featuring a model-generated summary
Aspect-based sentiment analysis
Domenic, Ray, Moaz and Hazel made up the aspect-based sentiment analysis team. Their prototype identifies positive and negative sentiments within documents and shows sentiment trends over time. This tool automates the manual consumption of vast amounts of data, while providing a deeper level of insight into sentiment about particular companies, products, and themes.
A graph from the aspect-based sentiment analysis demo showing sentiments across various industries by week
Kensho Crossover Services
Adam, Micahel, Justin, Drew and Jeremy on the Kensho Crossover Services team explored ways to integrate and layer existing Kensho products to create a compelling solution for users. Instead of accessing different individual solutions, users would be able to use and combine Kensho solutions in a single workflow.
A workflow showing possible ways to combine existing Kensho solutions
