Overview

In this project, I developed Trishul-AI, an AI-powered medical chatbot designed to provide basic health guidance and answer general medical queries. The goal was to build a safe and context-aware conversational system that leverages large language models and retrieval-based techniques to deliver relevant information while avoiding diagnostic or medical decision-making. By integrating LLM frameworks, vector databases, and scalable backend APIs, the system enables users to interact with medical knowledge in a structured and responsible way.

Key Features

• AI-Powered Medical Chatbot: Built a conversational assistant capable of answering general health-related questions using LLM-based reasoning and contextual understanding.
• Retrieval-Augmented Generation (RAG): Implemented a RAG pipeline where medical information is embedded and stored in a vector database, allowing the chatbot to retrieve relevant context before generating responses.
• Context-Aware Responses: Used semantic search with vector embeddings to ensure answers are relevant to user queries and grounded in retrieved knowledge.
• Safe Interaction Design: Implemented prompt engineering and safety constraints to ensure responses remain non-diagnostic and include appropriate medical disclaimers.

Technologies Used

• React and Next.js: For building the front-end codebase with clean, reusable components.
• Styled-Components: For managing styles dynamically using design tokens.
• GitHub Actions: For automating the pipeline and syncing design changes to the repository.
• Python & FastAPI: For building backend APIs and managing AI workflows.
• LangChain: For orchestrating LLM pipelines and implementing the RAG architecture.
• Pinecone (Vector Database): For storing embeddings and performing semantic similarity search.
• Next.js: For building the frontend user interface and handling user interactions.
• Clerk Authentication: For secure user authentication and session management.
• LLMs & Embeddings: For generating intelligent responses and understanding user queries.

Challenges and Learnings

One of the main challenges was ensuring that the chatbot provides helpful responses while maintaining safety and avoiding medical misinterpretation. This required careful prompt engineering, controlled knowledge retrieval, and proper response constraints. Another challenge involved optimizing the retrieval pipeline so that relevant context was consistently fetched from the vector database, improving the accuracy and reliability of generated responses.

Outcome

The Trishul-AI chatbot successfully demonstrates how LLMs and retrieval systems can be combined to create responsible AI assistants in sensitive domains like healthcare. The project strengthened my understanding of RAG architectures, vector databases, prompt engineering, and AI system integration, and showcases the potential of AI-driven conversational systems for providing accessible informational support.