ShalopAI

• 

  LLM Diagram 2

• 

  LLM Diagram 1

• 

  Result of chatbot work

• 

  Customer model

Inspiration

While designing this project we were greatly inspired by many of the current market leaders in the field of AI-assisted sales, such as Salesforce, Amplify.ai, and Twyla. Another huge source of inspiration for us was one of the sponsors of this event - Mercedes, who provided us with valuable insights and experiences from the automotive industry.

What it does

The project we made this weekend is a chatbot-consultant, which primary goal is to find an electric vehicle in assortment of Mercedes, that would be a perfect fit for a customer, based on its previous web behavior and metadata.

How we built it

After choosing carefully we decided to use Llama Index - a helpful and efficient Python NLP framework for our models build, and various datasets representing diverse buying behavior patterns of cars manufacturers' customers for training. The technology was then connected with a React frontend via convenient and reliable FastAPI for demonstrative purposes. We then fine-tuned the model and found a perfect architectural pattern for it to achieve the optimal performance.

Challenges we ran into

As mentioned earlier we are very thankful to the sponsors of this event for providing us with their technologies and experiences, which helped us to overcome the challenges of acquiring expert view on the problems of modern automobile industry.

Accomplishments that we're proud of

We are very proud of our team's overall performance and of competing in such an interesting use case with such greatly motivated peers. We are very thankful to the TUM.ai for that opportunity.

What we learned

Besides the powerful insights we gained about the industry, we were also able to practice our skills by working with state-of-the-art models and technologies, which has brought us a lot of valuable knowledge an hands-on experiences.

What's next for ShalopAI

The next steps of development for our project would be acquiring more training data from real market customers and increasing the number of model personalization parameters to deliver even better user experiences.

Built With

• javascript
• node.js
• python
• react