Prompt Engineer

We’re looking for a Prompt Engineer to own the language-model layer inside Admiral: the call intelligence that reads every marina phone call, the SMS and email automations that make sure no boater goes unanswered, and the systems that turn raw operational records into the numbers on an operator’s dashboard. The job is making LLMs boringly reliable in production — defining what correct means for each system, measuring it, and closing the gap. You should write with precision and think from the model’s side of the conversation.

What you’ll do

• Design: Architect the prompt systems, agent workflows, and conversation flows behind call intelligence, SMS and email automation, and marina reporting. Define the system prompts, few-shot examples, and guardrails — and the reasons behind each one.
• Evaluate: Build evaluation frameworks to measure prompt quality, accuracy, and reliability. Run systematic tests across edge cases, failure modes, and adversarial inputs. Use data to iterate, not intuition.
• Optimize: Improve latency, cost, and output quality against each other, on purpose. Model selection, structured outputs, retrieval — every trade-off documented, every regression caught by an eval before a marina sees it.
• Ship: Work with engineers to put prompts into production and own them there: deployment, monitoring, iteration. Your prompts don’t live in notebooks — they answer real phone calls at real marinas.

Preferred qualifications

• 2+ years of hands‑on experience with LLMs (GPT-4, Claude, Gemini, or similar).
• Deep understanding of prompt engineering techniques: chain-of-thought, few‑shot learning, function calling, structured outputs, and system prompt design.
• Experience building AI agents, chatbots, or automated workflows powered by language models.
• Strong analytical thinking with the ability to design and run rigorous evaluations.
• Proficiency in Python or JavaScript for prototyping and testing.
• Excellent writing skills — clarity, precision, and the ability to think from the model’s perspective.
• Familiarity with RAG architectures, vector databases, and embedding models is a strong plus.
• Experience in a startup or product‑focused environment preferred.