An AI-powered chatbot platform that leverages LLMs and RAG to provide intelligent customer support and solution recommendations in case of outages to ops teams.
Manhattan Co-Pilot/Maven is designed to address direct customer queries by integrating a chatbot with the customer-facing facade layer. The goal is to automate responses for common queries and reduce manual customer support intervention by 40%. This was achieved using a Retrieval-Augmented Generation (RAG) model, ensuring accurate and contextual responses.
Technology Stack:
Backend: Python (Django)
LLM Models: Gemini and OpenAI
Vector database : Pinecone
Frontend: Angular
Key Features:
Leveraged pre-trained LLM models Gemini and OpenAI GPT, eliminating the need for training on custom data to reduce unnecessary overhead.
Implemented the Chain of Thought approach to enhance the model's awareness of response quality and enable more precise decision-making.
Reduced manual intervention for generic and repetitive customer issues.
Implemented multilingual support to cater to a diverse customer base.
Leveraged Django Admin plugin to track customer feedback, distinguishing between good and bad responses. Multiple graphs were built using Grafana to visualize quality trends over time.
By combining an advanced LLM-powered chatbot with an efficient RAG-based approach, Manhattan Co-Pilot significantly enhances customer service efficiency while optimizing operational costs.