Sarah Mitchell
ML Engineer at Standard Compute
Sarah bridges the gap between ML research and production reliability. She spent five years at an NLP-focused startup building retrieval-augmented generation pipelines and hallucination detection systems before joining Standard Compute. Her writing focuses on the unglamorous, practical side of shipping AI — the architecture decisions that prevent models from confidently lying to your users.
Former NLP research engineer. Published work on grounded generation and factual consistency in long-context models. Holds a master's degree in computational linguistics.
Areas of expertise
LLM evaluation and hallucination mitigationRAG pipeline architectureMulti-pass inference workflowsPrompt engineering for production systems
Recent articles
My fix for hallucinating case notes was weirdly boring: stop stuffing context and split the job in two
2026-06-07 · 9 min read · Guide
My OpenClaw agent started writing nonsense and the real fix was a kill switch, not a better prompt
2026-06-07 · 11 min read · Guide
My agent remembered the whole meeting and still forgot the only parts that mattered
2026-05-18 · 9 min read · Guide
I thought llm tool calling would kill glue code and then my lights still wouldn’t turn on
2026-05-14 · 10 min read · Guide
Why does nobody talk about how expensive idle OpenClaw agents are?
2026-05-12 · 7 min read · Guide
I found the dumbest way to burn 500 LLM calls a day: polling an inbox every 5 minutes
2026-05-02 · 9 min read · Engineering
I thought ChatGPT Plus made OpenClaw unlimited and then my agent hit a wall overnight
2026-05-02 · 8 min read · Engineering