Deep dives, tutorials, and insights on building real-world AI systems.
Most RAG tutorials skip a critical production concern: what happens when sensitive documents enter your vector database? In this deep dive, I walk through building an automated PII scrubbing layer using spaCy NER that intercepts documents before they're embedded and stored — preventing names, emails, and medical IDs from surfacing in LLM responses.
A practical walkthrough of the hyperparameter choices, augmentation strategies, and architecture decisions that pushed my micro-expression classifier past the 85% barrier.
A step-by-step guide to packaging your trained PyTorch model in a Docker container, exposing a FastAPI inference endpoint, and deploying to AWS EC2.
Why MLflow is non-negotiable for ML projects and how to set it up in under 20 minutes. Covers experiment tracking, model registry, and artifact logging with real code examples.
Demystifying the confusion around LangChain's core abstractions — practical decision criteria for choosing between simple chains and autonomous agents for your AI application.