Inspiration
Witnessing the transformative power of artificial intelligence (AI) to streamline communication, resolve issues efficiently, and provide a seamless experience in a variety of platforms, we were motivated to create a platform that improves customer service operations, making the support process more personalized, effective, and responsive to customer needs.
All of us have experience using chat-bots on websites to do things such as check order status, or ask a general question about a product or service. More often than not, these have been poor experiences.
What it does
SupportFlow aims to solve the inefficiencies that exist in existing customer service platforms. When you start a new chat with SupportFlow, you're talking directly to our GPT-4 powered AI model. There are a number of scenarios where the AI model will successfully provide a solution to your inquiry, and will do so in a clear and concise manner without the need for a human representative. In cases where the model needs human assistance SupportFlow will transfer the chat session, including chat history and relevant customer information, to a human representative.
The human representative, or the support agent, will be able to provide immediate assistance to the customer without the customer needing to repeat any information they told the SupportFlow model. This is because the support agent's dashboard already has the complete chat history and customer information. The dashboard will inform the support agent why this chat was elevated, and offer possible solutions based on keywords from the chat history tied with organization-specific knowledge such as previously resolved chat sessions and product information.
How we built it
SupportFlow leverages OpenAI's API and function calling to conversationally handle user queries and perform the necessary actions to handle their requests and questions. In addition to the customer-facing chat UI, we also built an agent dashboard for when our AI needs to escalate a conversation to a human agent.
Challenges we ran into
- Reverse engineering a retail site's APIs to connect our functions to their data.
Accomplishments that we're proud of
- Building a faster, more natural chatbot experience than currently exists in many of today's companies
What we learned
- How to prompt a LLM to best decide when to call functions
- How to prompt an LLM to ask for help when needed
- How Home Depot secures its APIs using sessions
What's next for SupportFlow
- Shopify Integration to help more businesses improve their customer support
- Utilize OpenAI's whisper model to handle phone calls in addition to our chat


Log in or sign up for Devpost to join the conversation.