AI
Developed for a small finance bank, this email agent uses Llama3 and LangChain to analyze incoming emails. It identifies the user's intent, automatically creates a CRM ticket, notifies the assigned employee, and can even draft a reply by calling internal APIs to fetch necessary information.
This project was initiated to streamline customer support operations at a small finance bank by automating the initial stages of email-based queries.
The manual process of reading emails, identifying customer needs, creating CRM tickets, and routing them to the correct department was slow and prone to human error, leading to delayed responses.
The development involved fine-tuning the Llama3 model for banking-specific intents, building a robust LangChain pipeline for processing emails, and integrating with the bank's existing CRM and internal APIs.
Ensuring high accuracy in intent recognition for a wide variety of customer emails was challenging. This was addressed by augmenting the training data with synthetic examples and implementing a confidence-based fallback to a human agent.
The system was designed to be a "human-in-the-loop" agent. While it automates most of the workflow, it provides clear dashboards for human agents to monitor its activity and intervene when necessary.
The core of the agent is a LangChain pipeline that orchestrates multiple steps: email pre-processing, intent classification using Llama3, entity extraction for key information, API calls to the CRM, and generating a draft response.
The Intelligent Email Agent significantly reduced the manual effort required for handling customer emails, leading to faster response times and improved customer satisfaction.
The agent automated over 60% of incoming email workflows, allowing customer service representatives to focus on more complex and high-value interactions.
