Machine Learning in Production
I am not a researcher. I have spent my time on the other end of the pipeline: the part where a model has to serve traffic, stay within budget, and not embarrass anyone when it is wrong.
Scaling ML inference is the practical core of it — NVIDIA Triton, TorchServe, ONNX, compilation, GPU orchestration on Kubernetes. I first wrote it in 2022 and rewrote it in 2026, because the LLM era changed the cost arithmetic rather than the engineering. Self-healing techniques used by SRE runs the other way: machine learning pointed at operations, for forecasting, anomaly detection and risk classification.
Two posts deal with what happens when this goes wrong or goes fast. How the tech industry amplifies social stigmas with machine learning is the one I feel strongest about — bias enters at every stage of the pipeline, and shipping a model without understanding its dataset is negligence, not oversight. Agentic development looks at AI-assisted engineering with the same suspicion I would apply to any tool being sold to me as inevitable.
5 posts
How My 7-Year-Old Son Vibe Coded His Halloween Costume
Two weeks to build an ESP32 LED robot mask — 242 WS2812B LEDs, a 3D-printed see-through mesh, and a 7-year-old deciding what the faces should look like.
Agentic Development: Navigating the AI Revolution in Software Development
Exploring the rise of AI-powered development tools and their impact on software engineering, from GitHub Copilot to local LLMs. A deep dive into "vibe coding," agentic development, and how these tools reshape our workflows while examining opportunities and challenges.
How To Scale ML Inference to Improve Reliability, Speed, and Cost Efficiency
A practitioner's survey of the tools and strategies for scaling machine learning inference — covering NVIDIA Triton, TorchServe, ONNX, PyTorch compilation, OpenAI Triton, and GPU orchestration on Kubernetes.
How The Tech Industry Amplifies Social Stigmas with Machine Learning
Thoughts on the importance of applying Ethical and Responsible AI when designing software based on Machine Learning techniques.
Top 5 Machine Learning and Self-Healing Techniques used by SRE
Applying Machine Learning and Self-Healing techniques to the day operations of a production system has become common practices for most SREs. This post cover some real production use cases like automated failover, forecasting, anomalies detection, risk classification and so on.