AI Assist

Your Callcenter AI Assist: a customer support application to handle incoming calls, collect and analyze customer data, and match calls to the best employees based on their metrics powered by AI.

Comment 2

Inspiration

We were brainstorming about several projects for example: An AI Helper for people who can't read on write. This would be a good usecase for 3rd world-countries. In the end we decided to build an AI Assist for callcenters as we are annoyed to get useless support and waiting for hours in queues.

Our AI Assists uses "queue-theory" to match our staff with calls which are best to work on to reduce waiting time for all. Furthermore it helps agents to give better support. It allows managers and supervisors to find critical pain point of products but also to give feedback about general support success.

What it does

The software is basically a dashboard which has 3 different views. The calls view is the view of all incoming calls. The incoming calls are started by a bot which asks for a problem and the customerid to identify and map the customer to our platform. While the helpdesk staff is able to pick their best-match our software also proposes help for them. It matches problems described by the customers to our FAQ's using the Cohere API. Furthermore it also references to our problem database to make answering easy for the agent.

The software database model is pretty big. We use mock data to fill in some fields but also work with the Assembly AI API. We recorded some examples which are automatically deployed to a FTP server and sent to the API to give feedback about the clients problem. It provides also the highlights, the name of the customer, the customerid to map our customer database and more.

On the dashboard view, managers and supervisors can get critical business information. The software is also able to give a general overview of the sentiment of our clients. This will be a key factor to keep the business running and optimize critical structures.

A little example: Agents have more than 50 calls a day nowadays, they don't really have the time nor the energy to think about common problems. Our AI does that for you. We summarize everything and point you to crucial problems

In the history view, agents can check their history of calls. It also provides additional information to supervisors. They should be able to check team success and other details.

How we built it

Our database model is build with SQLalchemy based on Flask. Our backend also relies on Python and Flask. We use React with Typescript in our Frontend. The voice data is uploaded to as sever via FTP.

Challenges we ran into

We started with a pretty simple database model which then grew really fast. The relationship mapping brought us to some challenges. Another challenging part was the serialization of our database to json as we didn't use any module for that.

The challenging part of the frontend was to build a scalable architecture. Another important key is time management: some planning is nice but you're never know which little bug robs you several hours to fix it!

Accomplishments that we're proud of

Definitly our database model! We also like the outcome of our idea brainstorming as it tackles a real life problem! AI Assist DB Model

What we learned

  • Cohere API
  • Assembly API
  • SQLAlchemy
  • Database modeling
  • React/React Redux and Material Ui

What's next for AI Assist

  • In-depth market research, do potential customer interviews, and competitive analysis.
  • Write a business plan and greater pitch to outline our product (feature analysis,...), target market, financial projections. Potentially find investors and business partners.
  • Clean our code and develop an MVP
Share this project:

Updates

Submission history