Personalized virtual customer assistant

Workflow of the project represented in a flowchart

Inspiration

The torment of dealing with automated customer service was our main motivation. Most bots fail to give an intelligible answer that is related to the question asked rather than spitting out a pre-determined answer.

What it does

All-in-one virtual sales/customer assistant. It can help troubleshoot faulty products, in addition, to being able to sell and hagel with customers without being too "robotic". The bot figures out where the client is from how much are they willing to spend and how much can they spend. The estimations are used to create a virtual assistant of the same demographic as the client. In addition, it tries to verify the estimated information

How we built it

With front-end and back-end data management in addition to fine-tuning a pre-trained LLM. Then, integrate all of that with user data (acquired or estimated) in a prompt engineering scheme.

Challenges we ran into

We were dealing with a lot of technologies that we didn't have a lot (or any) experience with like back-end management and OpenAPI use. In addition, the process of reaching a functioning prompt-engineering scheme was labor-intensive