Pronab Chandra Roy

Architecture: This diagram details the end-to-end flow from data generation to real-time clinical visualization, highlighting the exact stream-processing mechanics.

• Data Quality: Handled out-of-order records using event-time watermarks, ensuring temporal accuracy in clinical events.
• Scalability: Processed 302,332 events at 1,000 events/sec with zero invalid records in the clean-run validation layer.
• Transferability: Architecture directly maps to real-world FHIR subscription feeds for live hospital integrations.

Architecture: This diagram breaks down the NLP pipeline, showing how unstructured clinical text is de-identified, parsed, and mapped to structured clinical phenotypes.

• Privacy-First: Implemented rigorous de-identification to safely handle Protected Health Information (PHI) in unstructured text.
• Clinical Context: Leveraged transformer-based NLP to extract complex medical entities from noisy, jargon-heavy discharge summaries.
• Interoperability: Mapped extracted entities to standard terminologies (ICD-10, SNOMED-CT), making unstructured notes available for SQL-based analytical querying.

Architecture: This diagram illustrates the medallion-style data engineering architecture (Bronze/Silver/Gold) used to transform raw ICU tables into ML-ready clinical datasets.

• Time-Series Aggregation: Engineered complex "first-24-hour" feature extractions from high-frequency ICU vitals (chartevents).
• Machine Learning Ready: Prepared baseline datasets optimized for in-hospital mortality prediction models.
• Secure Access: Built a guarded NL-to-SQL interface that translates natural language to SQL queries while preventing patient-level data exposure.

Architecture: This diagram shows the transformation of standardized CMS claim files into a dimensional model (Star Schema) for healthcare cost analytics.

• Claims Standardization: Parsed complex Medicare claim formats, linking beneficiary summaries to inpatient/outpatient/prescription events.
• Clinical Logic Implementation: Mapped ICD-10 diagnosis codes to Chronic Condition Data Warehouse (CCW) logic to identify patient chronic conditions.
• Predictive Analytics: Created "next-year high-cost" labels, enabling actuaries and researchers to train cost-prediction models.

Architecture: This diagram details the bioinformatics ETL process, ingesting raw genomic files and converting them into analysis-ready Parquet tables and graph databases.

• Multi-Omics Integration: Unified tumor genomic data (TCGA) with healthy baseline expression data (GTEx) for robust differential analysis.
• Bioinformatics Parsing: Processed complex Mutation Annotation Format (MAF) files and normalized raw RNA-seq expression counts.
• Graph Database Ready: Modeled genomic relationships (genes, mutations, patients, pathways) into node-edge CSV formats for Neo4j visualization.