Description
We're building Touly, an AI coaching product in the relationship and personal growth space. You can get a feel for where we're heading at touly.ai. The whole product rests on one thing: whether a large language model can hold a very specific, emotionally intelligent voice consistently across long, sensitive conversations. We're looking for a prompt engineer to own that layer. To be clear up front, this is a generative LLM role, not a data science or machine learning modelling role. If your background is in training predictive models, statistics, or classical ML, this isn't the right fit, and it saves us both time to say so now. The work. You'll write and iterate the system prompts that drive the core coaching conversation, and build the evaluation setup that tells us whether a prompt change improved or regressed behaviour rather than relying on gut feel. You'll tune how the model calibrates depth, holds a persona, and handles emotionally sensitive moments with care. You'll run structured comparisons across candidate models, and work closely with me, the founder, who owns the psychology and the voice, and with the development team on latency, prompt caching, and integration. What we need to see. Real hands-on experience prompting production LLMs such as Claude, GPT, or Gemini. Prompt work you've actually done, ideally a system prompt you can talk us through. A rigorous approach to evaluation: how you decide whether an output is good, and how you catch regressions. And the lang