From data analytics and applied machine learning to production-grade infrastructure — I build reliable software systems that turn complex data into actionable, everyday solutions.
Where Data Science meets Software Engineering
Building robust applications and performant backend APIs
Orchestrating, processing, and visualizing data at scale.
Translating complex data patterns into production-ready predictive models.
Provisioning reliable infrastructure and containerized deployments.
I’m a Software Engineer and Data Architect with 5+ years of experience building the systems that power modern AI. I specialize in turning complex data into hardened production environments—focusing on data integrity, inference efficiency, and scalable infrastructure. I build the engines that move a project from a successful experiment to a reliable product. When I’m not architecting systems, I document my engineering trade-offs and technical deep-dives in The Lab.
Designed a low-latency inference engine for a fintech client, serving complex ensemble models with sub-50ms latency requirements at high traffic.
Refactored a legacy cron-job system into a resilient Kafka + Airflow event-driven architecture. Zero data loss during live migration.
Diagnosed a 40% GPU utilisation bottleneck in a PyTorch training pipeline and resolved it through dataloader tuning and memory pinning.
Pickle is convenient, but it's a security nightmare and ties you to specific Python versions. Here is why ONNX and Protobuf are superior for production environments...
Cold starts are the enemy of real-time AI. We dive deep into the cost-benefit analysis of AWS Lambda vs. EKS for serving PyTorch models...
GPU utilization was stuck at 40%. The bottleneck wasn't the model, it was the data loading pipeline. Here is how we fixed it using num_workers and pin_memory...
I'm currently open to consulting opportunities and full-time roles in MLOps and System Architecture. If you have an interesting problem, I'd love to hear about it.
hello@devarchitect.devUsually responds within 24 hours.