Generative AI Conversational RAG Chatbot Software Solution

July 2024 - Present

Technologies Used: C#, .NET 8, Blazor, MudBlazor, Microsoft Azure, Semantic Kernel, CosmosDB

• Inherited the beta version of a RAG-Based LLM Chatbot Web Application, implementing new features and refining UI/UX for enhanced user interaction.
• Integrated image upload functionality and migrated the model from GPT-4 to GPT-4o, expanding the chatbot's capabilities to include visual question-answering and enhancing its overall versatility.
• Architechted a chat history feature with CosmosDB, utilizing efficient data structures to track message turns and minimize retrieval times, resulting in a more responsive and user-friendly interface.
• Developed a dynamic PDF access control system tied to RAG document corpus selection, implementing single-page chunking and display for critical documents to enhance security while maintaining relevant information accessibility.

Generative AI Conversational RAG Chatbot Software Solution

July 2023 - April 2024

Technologies Used: Python, JavaScript, ReactJS, FastAPI, Docker, Microsoft Azure, LangChain, Azure OpenAI, Azure AI Search

• Developed the FastAPI backend and ReactJS frontend full stack web application from scratch.
• Integrated Azure's cognitive search functionality, enabling document upload and search capabilities to work in tandem with LLM prompt engineering.
• Introduced real-time message streaming via WebSocket connections.
• Implemented authentication using Microsoft Azure Issued JWTs.
• Orchestrated the containerization of both backend and frontend using Docker and Microsoft Azure.

Capgemini Employee GenAI Solutions Hackathon

April 2024

Technologies Used: FastAPI, ReactJS, LangChain, Neo4j Aura DB, Cosmos DB, Azure AI Search, Azure OpenAI, Redis, Celery, Docker

• Implemented Azure AI Search to ingest and vectorize user documents, facilitating document retrieval based on semantic similarity from a vector index.
• Developed an asynchronous document processing system using Celery & Redis, allowing users to upload documents without disrupting the application's performance.
• Utilized an LLM to transform document chunks into a comprehensive knowledge graph, stored and managed in an Aurora Neo4j database.
• Designed and developed the complete system architecture from scratch, including the frontend, backend, and resource interactions. Implemented the end-to-end RAG flow, chatbot interactions, and message history to create a seamless user experience and efficient document retrieval process.