Project 14 · Cloud AI/ML
DE 1
StreamRx
Real-time Pharma Event Streaming
01 Abstract
Real-time Pharma Event Streaming
A high-throughput streaming pipeline that processes prescription and adverse-event data in real time over Kafka and Kinesis, with sliding-window pharmacovigilance signal detection.
02 Highlights
Methods & contributions
- Disproportionality metrics — Proportional Reporting Ratio (PRR), Reporting Odds Ratio (ROR), and Yates-corrected chi-squared with the standard delta-method PRR confidence interval.
- Signal thresholds — MHRA/EMA-style firing rules (PRR ≥ 2, χ² ≥ 3.84, ≥ 3 cases).
- Stream processing — Faust agents over Kafka and AWS Kinesis with sliding windows.
- Sink + infra — S3 Parquet sink and AWS MSK deployed via CloudFormation.
03 Architecture
Data flow
events → Kafka / Kinesis → Faust (sliding window)
→ PRR · ROR · χ² signal detector → S3 Parquet + alerts
04 Stack