For my IEEE-accepted research, “Hallucination Detection & Prevention in AI Coding Assistants,” I built a dual-stream RAG framework for C++ DSA code generation and a multi-signal verification approach. It reduced hallucination rate from 50.0% to 13.3% on 30 adversarial problems.
I also published a 50,728-row mental-health text dataset to the Hugging Face Hub and fine-tuned a DistilBERT classifier with SHAP/LIME interpretability. At NullClass, I worked as a Data Analyst Intern, using Python, SQL, and Power BI for analysis, visualizations, and dashboards.

