FINclusive

Inspiration

Many of us find it difficult to convey our thoughts during discussions for fear of being rejected. Some people gain conversational dominance by interrupting the speaker continuously or even confronting them, not allowing them to complete what they intend to share. This idea of FINclusive stemmed from such struggles many of us faced in our workplace. When we researched in-depth, we realized that many who try to gain conversational dominance do so because of biases. While some do it consciously, for many, it is their unconscious biases that come to play. Suppose there is someone or some app that can detect and remind me of my biases continuously. In that case, I will become more watchful of the before-mentioned situations and, over some time, I will overcome such biases due to reinforcement learning. That’s where the discussions on this idea started.

What it does

FINclusive is an app that detects unconscious biases in human-to-human conversations and alerts the people involved using a machine learning model. One technique of gaining conversational dominance, which is interrupting the speaker is addressed in FINclusive currently. Integrated with Finastra’s solutions, FINclusive can alert customer service representatives of such biases during conversations with customers, thus offering an enhanced inclusive customer experience.

How we built it

NLP is used for voice to speech conversion and AI model built on convolutional neural networks to identify biases. For this hack, Twitter sentiment analysis is used to identify interruptions. For speech recognition, we used webkitSpeechRecognition, though we intend to use NLP in our enhanced model. HTML5 and java script is used for UI. Consumer API of Fusion Fabric.cloud retrieves the customer information.

Challenges we ran into

• Unavailability of training data for the CNN model. Thus we used Twitter sentimental analysis library for this hack
• We intend to collect training data from movie dialogues and create a CNN model
• If organizations can share customer support recordings of call centers, that would be a great source of data for building the training model ## Accomplishments that we're proud of Excited that FINclusive can help several of us who have faced repercussions of unconscious biases in multiple ways. The potential end users of FINclusive will gain confidence over time, and will be devoid of their own unconscious biases, and well as others. Statistical reports and predictions on unconscious biases will help employees create an inclusive workplace. We believe FINclusive will be a huge value add to progressive organizations in driving their diversity and inclusion agenda.

What we learned

We learnt that technology can be used to uncover our unconscious biases. The possibilities with deep learning and neural networks are unlimited. With dedication, hard work and patience, we can define machine learning models that are helpful for human beings to reduce micro-stressors in their daily life.

What's next for FINclusive

Though the FINclusive demo addresses only one aspect of conversational dominance, which is interruptions, there is a huge possibility for incorporating areas such as detecting topic changes, sarcastic tones, confrontations etc. The solution also can be of use in meetings and discussions, where everyone is notified when conflicts arise or when one person is targeted. Discussions would become much more professional, eliminating workplace stresses. The same if deployed in customer support solutions will help create better inclusive customer experiences.

Built With

• express.js
• ffdc-consumer-(us)
• html5
• javascript
• node.js
• tensorflow
• twitter-sentimental-analysis
• webkitspeechrecognition