Engineering Playbook

Long-form engineering notes — each one a deep write-up. Newest first.

Evaluating Agent Reliability in Multi-Step Tasks→ Agents · Evals GPU Memory Optimization for LLM Serving→ CUDA · KV Cache From LoRA to Full Fine-Tune: When and Why→ PEFT · Training Prompt Engineering Patterns I Actually Use→ Prompting · CoT Designing Production RAG Systems That Don't Break→ RAG · Reliability Efficient Attention: FlashAttention to Multi-Query→ Memory · Transformers