CHINMAYEE GANGALAM

Built a multi-agent RAG system using LangGraph and LlamaIndex integrated with AWS Bedrock-hosted Meta LLaMA models to resolve HR payroll and benefits queries faster, reducing escalations to Tier-2 support. Fine-tuned LLaMA models using LoRA/QLoRA on proprietary payroll documentation and compliance datasets to improve domain-specific accuracy and reduce inference cost. Developed FAISS-based semantic retrieval for 500K+ internal HR policy and tax documents, achieving low-latency retrieval. Implemented LightGBM and Prophet/ARIMA forecasting for payroll volume and attrition trends, improving forecast accuracy. Engineered automated ETL pipelines from AWS Redshift into Tableau dashboards to surface real-time payroll anomalies and fraud indicators and reduce manual reporting effort.

Designed a hybrid CNN and BERT NLP pipeline in PyTorch for automated classification and NER on 50,000+ unstructured research documents, achieving strong F1 and reducing annotation time. Benchmarked XGBoost against deep learning baselines using structured research data stored in PostgreSQL, improving predictive accuracy and producing a peer-reviewed lab report. Prototyped an evaluation framework using GPT-3.5 Turbo and Claude for automated summarization and entity extraction, benchmarking against NLTK rule systems. Consolidated experimental data into Azure Data Lake, built a centralized warehouse, and created Power BI dashboards to visualize model performance metrics for faculty review.

Developed churn classification models using Python, Scikit-learn, and SVM for a retail vertical, achieving 87% accuracy and improving retention via targeted campaigns. Implemented model validation frameworks (cross-validation, confusion matrices) across production models to reduce overfitting and improve generalization accuracy. Used SHAP and LIME to explain model predictions to non-technical stakeholders and accelerate deployment approvals. Performed cohort analysis and data wrangling on 1M+ records to standardize inputs and reduce preprocessing time. Built Azure Data Factory ETL pipelines to automate weekly reporting. Built an NLP-based ticket classification module using Keras and NLTK and visualized trends in Power BI to improve routing accuracy.