Bhargav Peddi

Built real-time, scalable data pipelines with Spark, Kafka, Flink, Airflow, and Databricks to support tick-level, order book, and alternative datasets for quantitative research. Developed NLP workflows using transformer/LLM models to extract sentiment signals from SEC filings and earnings transcripts. Implemented forecasting and volatility models (ARIMA, GARCH, LSTM) and deployed ML-driven alpha strategies that improved Sharpe ratio, while also designing reinforcement-learning trading approaches and portfolio optimization models. Engineered low-latency execution components in C++ and Java and set up production MLOps with MLflow/Kubeflow, containerization, and monitoring.

Designed and delivered a clinical data lakehouse on Azure Databricks and Delta Lake to consolidate patient encounters, DICOM imaging metadata, and lab results, reducing reporting query latency. Built ELT pipelines with PySpark/Scala orchestrated by Airflow to standardize hospital source data into HL7/FHIR formats and map clinical codes, improving pipeline runtimes. Implemented near-real-time streaming with Kafka and Spark Streaming for remote telemetry ingestion, and added HIPAA-aligned governance with Azure RBAC, lineage tracking, masking, and encryption. Shipped FastAPI microservices exposing curated FHIR datasets with Kubernetes-based deployment, CI/CD, and observability via Prometheus/Grafana.

Created SQL and Python ETL jobs to consolidate sales, marketing, and inventory datasets into centralized reporting warehouses, reducing manual reporting effort. Built time-series and regression forecasting models in Python to improve forecast accuracy for planning, and designed/validated A/B tests using statistical hypothesis testing and cohort analysis to measure feature impact. Performed funnel analysis using SQL to surface conversion insights and built interactive Power BI dashboards tracking multiple KPIs. Contributed to a recommendation system using collaborative filtering to increase product suggestion click-through.

Developed a TensorFlow/Keras LSTM-based ensemble to forecast daily restaurant demand for a 7-day horizon. Engineered calendar and rolling temporal features and used time-ordered splits to prevent look-ahead bias. Visualized forecasts and residuals to support staffing and inventory planning, and used multiple seeded runs to stabilize evaluation metrics.

Created reproducible ML benchmarks comparing logistic regression, decision trees, and random forests across several classification and regression datasets. Built leakage-safe scikit-learn pipelines with preprocessing, engineered features, and grid-search tuning with cross-validation. Communicated results with EDA and diagnostic plots, and added automated tests to help ensure target leakage prevention and reliable metric outputs.