Machine Learning Engineer & Data Analyst building real-time data pipelines and validating ML models. (I verify every metric, from anomaly scores to parameter-efficient fine-tuning, before it ships.)
Real-time anomaly detection pipeline with 4.65ms avg latency on a 5.5% anomaly rate baseline. Verified performance: Sensor ROC-AUC 0.9952 / F1 0.8392 and Financial ROC-AUC 0.9893 / F1 0.7407. Uses Kafka, FastAPI, Isolation Forest + Autoencoder, and a WebSocket dashboard.
Calibrates model probabilities to ensure anomaly scores reflect true real-world likelihoods. Addresses the real operational cost of miscalibrated scores in production fraud pipelines.
End-to-end MLOps pipeline for Fraud Detection featuring automated CI/CD/CT, MLflow tracking, Evidently AI drift monitoring, and serverless AWS ECS Fargate deployment via CloudFormation.
End-to-end QLoRA fine-tuning pipeline adapting Mistral-7B-Instruct-v0.3 for closed-domain Medical Question Answering (MedQA). Demonstrates reproducible ML experimentation and quantitative before/after evaluation.
