Diparna Adhikary

• Validated and analyzed workforce and applicant-flow data using statistical methods and predictive modeling to support federal hiring compliance and audit reporting.
• Automated data integration across HR systems by building pipelines in Python, R, and SQL, significantly reducing manual reporting effort.
• Designed analytical dashboards and visual summaries to support leadership discussions on compliance, recruitment strategy, and workforce retention.

• Built 50+ interactive dashboards using Tableau and Power BI, enabling data-driven decision-making for 200+ stakeholders.
• Led client requirement-gathering sessions and translated business needs into scalable analytics solutions aligned with performance benchmarks.
• Automated business workflows using Power Automate and Power Apps, improving operational efficiency by ~20% while maintaining data accuracy.

• Resolved complex Power BI data access and reporting issues for sales and operations teams, reducing query resolution time by 25%.
• Managed dataset refreshes, report versioning, and documentation in Power BI Service to ensure data consistency and compliance.

• Designed and implemented an end-to-end machine learning system to forecast pricing adjustments for insurance products across diverse market segments, supporting data-driven pricing strategy.
• Engineered a high-dimensional feature set from multi-year market data, including lagged variables and interaction terms, and reduced it through correlation analysis and multicollinearity filtering.
• Built and deployed segment-specific regression models at scale, discovering that localized models significantly outperformed global approaches across heterogeneous product - market combinations.
• Delivered interactive visualizations and diagnostic tools to communicate model performance, market disruption effects, and forecasting insights to stakeholders.

Skills used: Python · Regression Modeling · Regularization (L1/L2) · Feature Engineering · Time-Based Analysis · Model Validation · Dimensionality Reduction · Data Visualization · Stakeholder Communication

Project details are intentionally high-level due to a non-disclosure agreement (NDA).

• Built a Retrieval-Augmented Generation chatbot over 63 MSU policy and regulatory PDFs (~2,186 pages), combining hierarchical chunking, semantic search, and cross-encoder reranking to surface precise, grounded answers with source citations.
• Engineered the pipeline with LlamaIndex (HierarchicalNodeParser), BAAI/bge-small-en-v1.5 embeddings, and ChromaDB for persistent vector storage; integrated open-source LLMs (Mistral-7B and Zephyr-7B GGUF) via llama-cpp-python and deployed a Gradio interface on Hugging Face Spaces with built-in SME feedback logging.

Skills used: Python · RAG · LlamaIndex · ChromaDB · Sentence Transformers · Cross-Encoder Reranking · LLMs (Mistral-7B / Zephyr-7B) · Gradio · Hugging Face Spaces

• Built an end-to-end machine learning pipeline to predict traffic accident severity using weather, temporal, and location-based features.
• Trained XGBoost models with experiment tracking via MLflow and deployed an interactive Streamlit app using Docker for real-time predictions.

Skills used: Python · XGBoost · Feature Engineering · Model Evaluation · MLflow · Docker · Model Monitoring

• Developed a retail analytics platform combining time series sales forecasting with customer sentiment analysis from reviews.
• Implemented Prophet and SARIMA models and deployed an interactive Streamlit dashboard to explore trends, forecasts, and business insights.

Skills used: Python · Time Series Forecasting · Prophet · SARIMA · NLP Sentiment Analysis

• Conducted exploratory analysis on historical flight and incident data to identify spatial and temporal patterns linked to the Bermuda Triangle.
• Designed interactive visualizations and deployed a Streamlit app to support data-driven storytelling and exploration.

Skills used: Python · Exploratory Data Analysis · Data Visualization · Geospatial / Temporal Analysis