This project involved:  Extracting text from pdf’s and pre-processing the data in the pdf in the form of tables and graphs.  OpenaAI’s text-embedding-3-large model to embed the text from pdf files to ElasticSearch Vector Database.  Generating contextualized query by passing the chat_history and user query to the LLM i.e. OpenAI’s GPT-4o.  Retrieving the vectors from the vector database on the basis of contextualized query.  Passing the chunks retrieved along with chat_history and user query to the LLM to form the responses.  Saving the responses and the user query in MongoDB on the basis of the user’s email Id.  The ElasticSearch and MongoDB databases are set in the server in the docker container.  Build

This project involved the development of a voice-enabled conversational AI agent that simulates the behavior of human agents in data collection scenarios, such as those found in call centers. The system is designed to gather structured user information through a guided interaction while dynamically responding to user inputs. Key Features:  Interactive Data Collection Flow: The agent initiates and manages a multi-turn conversation to collect specific user information based on a predefined question set.  LLM-Based Response Interpretation: A large language model (LLM) is used to analyse user responses in real-time. It determines whether the input is: o A valid answer to the current question o A query about the process o A refusal to answer o A request for clarification  Intent Classification and Dynamic Handling: The LLM classifies the user’s intent and dynamically adjusts the conversation flow. It can: o Provide additional explanation if the user asks for clarification o Rephrase questions if needed o Acknowledge and respond appropriately to refusals o Continue smoothly with the next question when appropriate  Retrieval-Augmented Generation (RAG) Integration: If a user requests specific information beyond the scope of basic data collection (e.g., FAQs, procedural details), the agent intelligently routes the query to a RAG system to retrieve and present accurate and contextually relevant answers.  Database Connectivity: The agent is connected to a persistent storage system, enabling it to: o Store newly collected user data o Retrieve and pre-fill previously collected data o Update or replace user data based on new inputs  Intelligent Routing Engine: A decision-making mechanism determines in real-time whether to: o Continue with the next step in the data collection flow o Route the query to the RAG system o Interact with the database for read/write operations