AI-Powered Customer Support Chatbot

Engineered an intelligent, end-to-end AI chatbot solution that provides instant, accurate answers from diverse company documentation, significantly boosting customer satisfaction and reducing support costs.

Context

The primary need was to enhance customer support efficiency and internal knowledge sharing through an intelligent chatbot system capable of interacting with a wide range of company documents. Providing quick, precise answers to complex questions based on extensive documentation was time-consuming for the support team, and employees struggled to autonomously find necessary information.

Project Objectives

• Deliver rapid and accurate answers to complex inquiries by leveraging a vast documentation base.
• Reduce the workload on the human support team, allowing them to focus on more complex issues.
• Improve internal productivity by providing employees with easy and autonomous access to necessary information.
• Ensure 24/7 availability of customer support.
• Develop a scalable and reliable chatbot infrastructure.

Implemented Solution

I designed and implemented an end-to-end intelligent chatbot RAG solution with the following key features and components:

Key Steps

• ETL Pipeline (Python): Developed a robust pipeline for extracting, transforming, and loading data from diverse sources (documents, databases, APIs) into vectorized embeddings.
• Vector Database (ChromaDB): Indexed documents using embeddings for efficient semantic search and retrieval of relevant information.
• Metadata Indexing: Enriched documents with comprehensive metadata to enable precise, contextual filtering of information, improving answer accuracy.
• Large Language Model (GPT-4o): Leveraged OpenAI's advanced GPT-4o model for generating highly relevant, natural, and coherent responses.
• LLM Abstraction Framework (LlamaIndex): Utilized LlamaIndex to seamlessly connect the chatbot to the vector database and the LLM, enabling intelligent querying and response generation.
• Salesforce API Integration: Implemented automated ticket creation in Salesforce when the chatbot is unable to resolve an inquiry, ensuring no customer issue is left unaddressed.
• Session Memory (Redis): Integrated Redis for conversational memory, allowing for more contextualized and personalized interactions across longer conversations.
• REST API (FastAPI/Python): Developed a high-performance REST API using FastAPI and Python, serving as the backend for the chatbot and managing document context.
• Performance Monitoring (Phoenix by Arize AI - Self-hosted): Set up a self-hosted Phoenix instance for continuous tracking and optimization of chatbot accuracy, latency, and overall performance.
• Deployment (AWS EC2 with Docker): Deployed the entire infrastructure on AWS EC2 using Docker for containerization, ensuring a scalable, reliable, and easily manageable environment.
• WordPress Integration (PHP Plugin): Developed a custom PHP plugin for seamless integration of the chatbot widget directly into the WordPress website, ensuring easy adoption and accessibility.

Skills Used

• Data Processing: Python, pandas
• Vector Database: ChromaDB
• LLM Framework: LlamaIndex
• Large Language Model: OpenAI GPT-4o
• Backend/API: FastAPI, Python
• Session Memory: Redis
• CRM Integration: Salesforce API
• CMS Integration: WordPress, PHP
• Monitoring: Phoenix by Arize AI
• Deployment: Docker, AWS EC2

Outcomes

• Significant Improvement in Response Times: Customers receive instant answers to their inquiries, drastically reducing wait times.
• Increased Customer Satisfaction: Achieved higher customer satisfaction due to 24/7 availability of rapid and accurate responses.
• Reduced Workload on Human Support Team: The chatbot successfully handled a significant portion of routine inquiries, allowing human agents to dedicate more time to complex and high-value issues
• Enhanced Internal Productivity: Employees gained immediate and autonomous access to necessary company information, streamlining internal processes and reducing time spent searching for answers.
• Scalable Customer Assistance: The solution provides scalable customer support, capable of handling fluctuating inquiry volumes without increasing human resource demands.
• Continuous Optimization: Ongoing performance monitoring ensures the chatbot's accuracy and effectiveness are constantly refined, leading to sustained value.