- Published on
How to tell whether an LLM system works, and how to keep it working: test cases, graders, judges, groundedness, CI gates, live signals, traces, latency, schemas, injection defense, and PII masking.
Agents
All Write-ups
- llm (10)
agents (7)
- ai (5)
- guide (5)
- infrastructure (4)
- deep-learning (3)
- gpu (3)
- inference (3)
- loop-engineering (3)
- mlops (3)
- python (3)
- pytorch (3)
- continual-learning (2)
- data-science (2)
- distributed-training (2)
- evals (2)
- guardrails (2)
- kv-cache (2)
- networking (2)
- reinforcement-learning (2)
- anyscale (1)
- architecture (1)
- attention (1)
- azure (1)
- causal-inference (1)
- claude-code (1)
- containers (1)
- context-engineering (1)
- cuda (1)
- data-engineering (1)
- devops (1)
- distributed-computing (1)
- docker (1)
- experimentation (1)
- feature-engineering (1)
- graph-engineering (1)
- grpo (1)
- hardware (1)
- harness-engineering (1)
- hooks (1)
- jepa (1)
- kernel (1)
- linux (1)
- machine-learning (1)
- mathematics (1)
- ml-systems (1)
- model-based-rl (1)
- moe (1)
- multi-agent (1)
- number-theory (1)
- numpy (1)
- observability (1)
- pandas (1)
- particle-physics (1)
- physical-ai (1)
- physics (1)
- primes (1)
- quantum-mechanics (1)
- ray (1)
- reasoning (1)
- riemann-hypothesis (1)
- rlvr (1)
- robotics (1)
- self-improving-agents (1)
- skills (1)
- standard-model (1)
- statistics (1)
- subagents (1)
- system-design (1)
- systems (1)
- tpu (1)
- training (1)
- vllm (1)
- world-models (1)
- Published on
An agent that gets better on its own has to change something about itself, and something has to check the change. This post walks the six things it can change, the checker that decides whether any of it worked, the ceiling it runs into, the ways it breaks, and the loop teams actually run in production.- Published on
Forty production questions on agentic systems, answered plainly: evals, guardrails, quality measurement, observability, security, and token economics.- Published on
What graph engineering is, how it differs from loops and workflow engines, when to build one, a refund approval worked through end to end, and the failure modes to watch for.- Published on
Why AI work moved from single prompts to loops, the main types of loops, and examples of goal-driven loops in practice.