Short bio
Senior AI Engineer with 8 years delivering production machine learning and deep learning solutions across startup and enterprise environments, including regulated and high-growth 0-to-1 projects. Hands-on with Python, TensorFlow, PyTorch, scikit-learn, time series analysis and computer vision, and experienced in privacy-first techniques like differential privacy and federated learning. Proven at building anomaly detection, smart alerting and model explainability with MLflow, ONNX, SHAP, secured deployments on AWS, Kubernetes, Docker, and compliant pipelines for HIPAA/GDPR audits while collaborating cross-functionally to ship scalable ML products.
Tech stack
Work experience
- Senior AI EngineerJune 2023 – Present3 y.
Architected time series and computer vision pipelines using Python, PyTorch, and TensorFlow to support healthcare monitoring pilots, improving anomaly detection recall by 24% across 3 trusts.
Responsibilities:- Designed privacy-preserving model workflows with differential privacy and federated learning, eliminating centralized PHI exposure in 2 production pilots and meeting client compliance objectives.
- Integrated model inference into production services using Docker, Kubernetes, and AWS SageMaker, reducing deployment time per model from days to under 2 hours.
- Implemented smart alerting and notification systems using Kafka, Prometheus, and rule orchestration to decrease clinician response time by 33% on monitored cohorts.
- Developed hybrid LSTM-CNN anomaly detection models in PyTorch, achieving 91% detection accuracy on multisensor datasets exceeding 1M records and lowering false alarms by 29%.
- Orchestrated MLOps with MLflow, GitLab CI, and Terraform, cutting model retraining cycles from 7 days to 18 hours while maintaining reproducible experiments.
- Validated model outputs and maintained labeled data and metadata in PostgreSQL and S3, establishing auditable data lineage for regulatory reviews across 2 clients.
- Coordinated cross-team integration of REST and gRPC inference endpoints into mobile and web applications, reducing end-to-end latency by 45% in production tests.
- Spearheaded compliance and security workstreams to enforce HIPAA, GDPR, and encryption-at-rest via KMS, producing audit-ready documentation for 100% of deployed models.
- Mentored 4 ML engineers on model explainability and operationalization using SHAP, LIME, and CI best practices, raising team delivery velocity by 40% over 6 months.
- Full Stack EngineerApril 2020 – April 20233 y.
Engineered NLP and document classification pipelines with Python, PyTorch, and spaCy to extract legal events from 10M+ documents, increasing event F1 by 18% on priority labels.
Responsibilities:- Implemented privacy-focused redaction and differential privacy measures to de-identify client PII, ensuring GDPR compliance across 2TB of case data and reducing review time by 22%.
- Constructed user-access time series anomaly detection using LSTM models on access logs, lowering anomalous access incidents by 47% across internal systems.
- Automated CI/CD and model deployment using Docker, Kubernetes, and GitHub Actions, decreasing release-to-production time by 60% and supporting a 99.9% uptime SLA.
- Tuned transformer-based classifiers to 95% precision on priority document types using Transformers and transfer learning on a 200k sample set.
- Reduced inference latency by 50% via ONNX conversion and GPU/CPU routing policies using CUDA and AWS EC2, enabling real-time document scoring under 120ms.
- Collaborated with product, legal, and QA teams in agile sprints to design smart alerts for case escalations using Kafka and Prometheus, cutting detection-to-action time by 35%.
- Delivered secure data ingestion and storage pipelines with PostgreSQL, Redis, and S3, ensuring encrypted-at-rest storage and schema versioning for compliance audits.
- Established model monitoring with Prometheus, Grafana, and drift detection that triggered automated retraining when concept drift exceeded 5%, reducing silent failures by 83%.
- Documented model cards, runbooks, and compliance artifacts to support internal and external audits, shortening regulator review cycles by 14 days.
- Software EngineerApril 2018 – January 20202 y.
Built fraud detection prototypes with Python and scikit-learn leveraging transaction time series, reducing false positives by 22% on a 5M transaction dataset.
Responsibilities:- Led streaming ingestion using Kafka, Spark, and Airflow to process 20M events per day for real-time analytics and risk scoring.
- Improved model performance 15% through feature engineering and rigorous cross-validation, persisting features in PostgreSQL and Redis for low-latency lookup.
- Analyzed mobile deposit computer vision use-cases with OpenCV and TensorFlow Lite, achieving 89% recognition accuracy in pilot releases.
- Migrated legacy ETL to containerized microservices using Docker and Kubernetes, lowering infrastructure costs by 18% and enabling horizontal scaling.
- Refactored monolithic services into RESTful Node.js APIs, increasing deployment frequency from monthly to weekly and improving observability.
- Streamlined Incident Management and Change Management aligned with ITIL practices, reducing MTTR by 28% and improving post-incident review completion rate.
- Supported cross-functional teams to integrate models via gRPC and REST into web and mobile apps, maintaining 99.5% service availability during peak loads.
