Sai Mahanth

Built an end-to-end ELT pipeline on Snowflake using dbt, implementing a medallion architecture (bronze/silver/gold) to transform raw Airbnb data into analytics-ready models. Designed fact tables and dimensional models in the gold layer, with SCD Type 2 history tracking via dbt snapshots.

Built and evaluated a multi-stage RAG pipeline over 25 resumes (LangChain, OpenAI API); ran systematic error analysis to diagnose retrieval failure modes, then benchmarked fixes that improved retrieval precision by an estimated 30%. Implemented structured, grounded output with Pydantic schemas and validation checks that reduced hallucinated answers by an estimated 40%.

Built parameterized ETL pipelines in Azure Data Factory ingesting EMR data from Azure SQL to ADLS Gen2 Bronze/Silver/Gold layers (5+ hospital sources); implemented incremental load strategies with watermark triggers reducing processing time by 40%. Developed PySpark notebooks for data cleansing and standardization of medical codes (ICD-10, CPT, HCPCS).

Built end-to-end streaming pipeline using Azure Event Hubs ingesting ride events into Databricks Lakeflow with bulk loads into ADLS Gen2; architected Medallion Lakehouse supporting high-volume real-time ingestion, processed 15,000+ ride events per day with sub-2-minute end-to-end latency. Designed Gold-layer star schema using Databricks CDC flows with SCD Type 1/2 dimensions.

Designed scalable Azure Data Factory ETL pipelines to process 15+ years of UN Comtrade petroleum trade data across 100+ countries; applied forecasting and network analysis, achieving an estimated 85% forecast accuracy (MAPE under 15%) in modeling regional trade flows through 2028.

Engineered aerospace data pipeline ingesting 2M+ simulation records; removed 35% duplicate entries through custom PySpark validation framework, improving model training accuracy by 18%. Designed dimensional data models and schemas enabling 3 cross-functional teams to run performance benchmarking analysis independently.

Architected real-time user activity pipeline (Python, Apache Kafka) ingesting 50K+ events/day; dual-storage strategy enabling analytics on 6M+ event records with sub-100ms query latency. Built 5+ production ETL/ELT pipelines (Azure Data Factory, Airflow) implementing Medallion architecture and data validation checks, reducing downstream errors by 40%.

Pursued a Master of Science in Data Science, focusing on data engineering and analytics.

Earned a Bachelor of Engineering in Mechanical Engineering.