Shrishti Karhana

I’m an aspiring Data Scientist with internship experience across LLM response evaluation, multilingual NLP, and time-series forecasting. I focus on turning messy data into dependable models, clear metrics, and user-ready insights.

In my Data Science internships, I processed 3+ years of electricity market data using Python and SQL, engineered lag and rolling features, and built ARIMA, Random Forest, and XGBoost forecasting models. Through feature engineering and tuning, I reduced MAPE from 12% to 4.8%, and I delivered forecasting insights with Streamlit dashboards integrated with Tableau.

For LLMs, I evaluated and ranked 1200+ responses to improve output quality, reasoning, and alignment—spotting hallucinations, ambiguity, and safety issues, then annotating datasets with structured feedback. I’ve also built an offline speech translation pipeline (ASR → MT → TTS) for Hindi–Tamil–Telugu using IndicTrans2, and I’m energized by work that improves reliability from end to end.