Hi, I'm Ankita 👋
Data Scientist & AI Engineer building LLM-powered systems and production data pipelines. MS in Data Science @ Indiana University Bloomington.
I like taking messy, real-world data (federal incident records, live transaction streams, philanthropic grant records) and turning it into systems that actually hold up when you check the numbers, including the numbers that aren't flattering.
🔧 What I work with
Python SQL DSPy Chroma LangChain LlamaIndex Hugging Face Transformers MCP Scikit-learn SHAP MLflow Apache Airflow dbt Apache Kafka Apache Spark Power BI Tableau R Causal Inference
📌 A few projects
Real-Time Transaction Fraud Detection Pipeline Kafka + Spark Structured Streaming detecting anomalous transactions, with NVIDIA NIM-generated explanations and account risk status exposed as an MCP tool. Found and fixed a flawed evaluation methodology mid-project, the honest results (and the fix) are documented in the README.
Natural Gas Pipeline Risk Mapping Geospatial ML predicting high-severity pipeline incidents from PHMSA data. Caught a data leakage bug that was inflating ROC-AUC to 0.998, fixed it, and shipped an honest 0.842 with SHAP explainability instead.
US Domestic Flight Delay Analytics Pipeline End-to-end ELT pipeline (Airflow, dbt, PostgreSQL, Docker) ingesting 500K+ flight records into a 3-layer analytical warehouse.
Online Retail Customer Analytics RFM segmentation and cohort analysis on UK retail data, with a repurchase-risk model flagging high-value, high-risk customers.
📫 Connect ankinaik@iu.edu