Explore projects

Harry Potter RAG Chatbot is an AI-powered question-answering application built using Streamlit, FAISS, SentenceTransformers, and Groq. The system uses Retrieval-Augmented Generation (RAG) to retrieve relevant information from Harry Potter text datasets and generate accurate, context-aware responses. It performs semantic search using vector embeddings and FAISS to find the most relevant content before sending it to the language model. The chatbot provides an interactive user experience and is deployed on Hugging Face Spaces for easy accessibility.

Our IPL 2026 Assistant is a RAG-based application designed to provide accurate, fact-based insights from fragmented match data. By grounding user queries in our official matches.csv records, the system eliminates model hallucinations regarding IPL 2026 statistics, venues, and team performance.

Technical Implementation:

Knowledge Base: We transformed structured CSV match records into natural-language sentences, significantly improving semantic retrieval quality.

Retrieval Strategy: The system utilizes all-MiniLM-L6-v2 embeddings and FAISS to facilitate natural language questioning, allowing users to query trends like venue-specific performance efficiently.

Generation: The pipeline retrieves the top 2–3 relevant match documents and processes them through a strictly grounded system prompt to synthesize clear, professional, and factual answers.

Running the Application:

Activate Environment: Run .\venv\Scripts\activate in your terminal.

Launch: Execute streamlit run app.py to start the interface.

Access: View the assistant at http://localhost:8501 in your browser.

This design ensures high domain relevance and low latency, perfectly suited for sports enthusiasts requiring reliable data analysis on the go.

RAG-based AI chatbot using Streamlit

My project for the AI chatbot as assigned by Swecha during my internship

Developed a PDF-based RAG (Retrieval-Augmented Generation) chatbot designed to answer college-related queries efficiently using document-based semantic search. Used college.pdf as the primary knowledge source containing attendance policies, semester examination details, placement eligibility, academic guidelines, and extracurricular club information. Selected PDF documents because educational institutions commonly store notices, rules, and handbooks in PDF format, making the chatbot practical for real-world applications. Implemented PDF loading using PyPDFLoader to extract textual content and convert it into readable document objects for further processing. Applied text chunking using RecursiveCharacterTextSplitter to divide large document content into smaller manageable sections for better processing. Configured chunking with chunk_size=500 and chunk_overlap=50 to improve retrieval performance and maintain contextual continuity between chunks. Used overlapping chunks to prevent context loss when important information is split across multiple sections. Improved semantic retrieval accuracy by enabling the chatbot to understand meaning rather than relying on exact keyword matching. Integrated the Hugging Face embedding model sentence-transformers/all-MiniLM-L6-v2 for generating high-quality vector embeddings. Chose the embedding model because it is lightweight, fast, beginner-friendly, and efficient for semantic search tasks. Converted textual content into vector representations, allowing similar meanings to be matched effectively during user queries. Enabled intelligent query handling where questions with different wording could still retrieve relevant answers from the document.