Charanpreet Singh

I've built end-to-end machine learning systems for customer churn prediction at Celebal Technologies and air-quality forecasting at the Punjab Pollution Control Board.

For churn prediction, I developed preprocessing, exploratory analysis, feature engineering, supervised training, evaluation, and a Streamlit dashboard for real-time churn inference. For AQI forecasting, I used SARIMAX on 2018–2023 air-quality data and achieved RMSE values of 8.77 for PM2.5 and 17.67 for PM10.

My projects include Perception-HUB, a containerized FinBERT sentiment-analysis microservice fine-tuned with LoRA, achieving 97% accuracy and a 0.826 F1 score. I automated Docker deployment through AWS CodePipeline and ECS, reducing manual intervention by 40%.

I've also built RAG systems combining vector embeddings and knowledge graphs, using LangChain and ChromaDB to improve context grounding and reduce irrelevant responses. My IEEE conference paper focuses on mitigating intrinsic hallucinations in large language models through orthogonal fine-tuning and Dirichlet prior calibration.