CDA: A Conversational Data Analyst System Using Generative AI, LangChain, and Natural Language Processing

Authors:

Mrs. D. Gayatri

Karakavalasa Yaswanth Kumar, Gedela Uday Kiran, Gulla Hari Krishna, Gollu Sai Krishna

Dept. of Information Technology MVGR College of Engineering (A) Vizianagaram, Andhra Pradesh, India

Abstract—Structured data has grown faster than most organizations can analyze it, and the tools built to handle that gap

• Excel, Power BI, Tableau still require more technical skill than most people have. This paper describes the Conversational Data Analyst (CDA), a full-stack system that lets users query structured datasets in plain English, without writing code or learning specialized CDA pairs a React/Next.js frontend with a FastAPI backend and uses LLaMA 3 70B through LangChain’s agent framework to interpret queries and generate Pandas-based analysis on the fly. The system renders charts in real time via Recharts and Matplotlib, exports PDF reports through FPDF2, and delivers analysis summaries by email in the background. All data stays local — nothing goes to external cloud storage. Session state is kept in SQLite via SQLAlchemy, so users can ask follow-up questions without losing context. In testing, the system answered approximately 95% of standard queries correctly and 88% of complex multi-step ones, with response times under 5 seconds on datasets of up to 21,613 rows.

Index Terms—Conversational AI, Natural Language Processing, Data Analytics, Large Language Models, Generative AI, FastAPI, LangChain, LLaMA, Data Visualization, Report Generation, Pandas, Full-Stack System