Inference Pipeline

Context

In a previous job, I was tasked with detecting anomalous behavior based on the readings of multiple sensors. Previously, a heuristic was used that would trigger alerts if a threshold was met. I built, prototyped, and deployed a real-time machine learning inference pipeline using our IoT provider and AWS services. I also implemented automatic model retraining on a daily schedule.

Impact

• Developed a POC and deployed it to a construction site
• Detected all leaks in testing dataset at least 50% sooner than the threshold method.
• Presented solution to CEO and gained approval for further development

Skills Used

• Labeled data collection
• AWS: Glue, Kinesis, API Gateway, DynamoDB, Lambda, SNS, SageMaker
• ETL, Exploratory Data Analysis
• Apache Spark, Kafka
• Machine Learning
• IoT
• Webhooks
• SQL
• Python
• Stake
• PowerBI
• Snowflake