Web app with content
To any financial institution, the most valuable asset to increase revenue, remain competitive and drive innovation, is aggregated market and client data. However, a lot of data and information is left behind due to lack of structure.
So we asked ourselves, what is a source of unstructured data in the financial industry that would provide novel client insight and color to market research?. We chose to focus on phone call audio between a salesperson and client on an investment banking level. This source of unstructured data is more often then not, completely gone after a call is ended, leaving valuable information completely underutilized.
What it does
Structerall is a web application that translates phone call recordings to structured data for client querying, portfolio switching/management and novel client insight. Structerall displays text dialogue transcription from a phone call and sentiment analysis specific to each trade idea proposed in the call.
Instead of loosing valuable client information, Structerall will aggregate this data, allowing the institution to leverage this underutilized data.
How we built it
We worked with RevSpeech to transcribe call audio to text dialogue. From here, we connected to Microsoft Azure to conduct sentiment analysis on the trade ideas discussed, and displayed this analysis on our web app, deployed on Azure.
Challenges we ran into
We had some trouble deploying our application on Azure. This was definitely a slow point for getting a minimum viable product on the table. Another challenge we faced was learning the domain to fit our product to, and what format/structure of data may be useful to our proposed end users.
Accomplishments that we're proud of
We created a proof of concept solution to an issue that occurs across a multitude of domains; structuring call audio for data aggregation.
What we learned
We learnt a lot about deploying web apps, server configurations, natural language processing and how to effectively delegate tasks among a team with diverse skill sets.
What's next for Structurall
We also developed some machine learning algorithms/predictive analytics to model credit ratings of financial instruments. We built out a neural network to predict credit ratings of financial instruments and clustering techniques to map credit ratings independent of s_and_p and moodys. We unfortunately were not able to showcase this model but look forward to investigating this idea in the future.