Practical notes for building reliable AI systems: LLMs, retrieval, data stores, evaluation, MLOps, observability, and system design tradeoffs.
12-stage path from Python async to production GenAI — FastAPI, RAG, agents, inference, safety, and ship.
Transformers, prompting, retrieval, agents, cost control, and reliability — with trade-offs at every layer.
PostgreSQL, Redis, pgvector, and retrieval stores — schema design, indexing, and production query patterns.
CI/CD for ML, drift detection, model serving, governance, and the operational practices that prevent 3am pages.
Prompt versioning, RAG operations, LLM evaluation pipelines, fine-tuning workflows, and guardrails.
Production AI architecture and trade-offs: reliability, latency, scale, and the decisions that define a system.