Prompt Engineer

You will design, test, and iterate prompts and interaction patterns for LLM-powered features. The focus is on improving output quality, consistency, safety, and cost-efficiency through structured prompting, evaluation, and experimentation.

The core responsibilities for the job include the following:

Prompt Design and Iteration:

• Create structured prompts, templates, and system instructions for LLMs (GPT, Claude, Gemini, and open-source models).
• Apply best practices: context management, instruction hierarchy, few-shot examples, and tool/function calling concepts.
• Test and debug prompt behavior; analyze failures, reduce hallucinations, and improve consistency.

Evaluation and Quality:

• Define evaluation criteria and rubrics; build test sets and human review workflows.
• Track quality metrics: task success, groundedness, hallucination rate, completeness, safety, and user satisfaction.
• Run A/B-style experiments to measure improvements over time.

PromptOps and Governance:

• Version-control prompts; document assumptions, edge cases, and limitations.
• Optimize for token efficiency, latency, and cost; implement change management and release processes for prompts.

Collaboration:

• Work with product, engineering, and domain experts to translate business needs into prompt requirements.
• Partner with UX to design intuitive AI interactions, guardrails, and safe outputs.

Requirements:

• Strong written communication; ability to write clear, unambiguous instructions.
• Working knowledge of LLMs, their failure modes (hallucinations, bias), and prompting techniques (zero-shot/few-shot, CoT alternatives).
• Structured thinking; ability to evaluate AI outputs systematically and iterate.
• Basic Python or scripting skills for automation and evaluation (nice-to-have but valuable).

Nice-to-Have:

• Experience with RAG, retrieval strategies, or integrating prompts into applications.
• Familiarity with evaluation tooling (e. g., MLflow, custom eval scripts) and prompt versioning.
• Exposure to fintech, payments, or regulated use cases.