Assistant Prompt Engineering Jobs in New York (NOW HIRING)

LLM Interaction Design / Prompt Optimization / GenAI Application Quality

Role purpose

Design, test, govern, and continuously improve prompts, system instructions, conversation flows, and interaction patterns for AIRP LLM applications and related citizen-development experiences. The role ensures model outputs are accurate, grounded, safe, consistent, cost-aware, and aligned with business and compliance expectations.

Client-specific emphasis

• Prompt work must support enterprise business use cases, not generic chatbot experimentation.
• Reusable prompt patterns should be suitable for AIRP and, where applicable, Copilot Studio / Power Platform citizen-development scenarios.
• Candidates must understand prompt security, sensitive data handling, citations/grounding, and structured evaluation.

Primary ownership

• Prompt patterns, system instructions, response templates, and conversation policies for AIRP LLM use cases.
• Prompt testing, versioning, evaluation, and quality-improvement workflows.
• Reusable prompt libraries and guardrail patterns for business teams and responsible citizen development where applicable.

Key responsibilities

• Design prompts for chatbots, copilots, RAG systems, document analysis, summarization, workflow agents, knowledge assistants, and decision-support experiences.
• Develop system prompts, few-shot examples, tool-use instructions, response formats, escalation logic, citation behavior, and conversation policies.
• Optimize prompts for KYC support, credit underwriting support, governance tracking, pitch book generation, Banker 360, Customer 360, deal library intelligence, financial crime quality, and sanctions screening use cases.
• Build reusable prompt libraries and templates aligned to enterprise standards, business domains, AIRP patterns, and citizen-development guardrails.
• Evaluate prompt performance using metrics such as task success, groundedness, hallucination *, completeness, safety, user satisfaction, latency, and token cost.
• Partner with engineers to implement prompt versioning, testing, deployment, and monitoring in production systems and CI/CD workflows.
• Support RAG quality by assessing retrieval context, chunking quality, source citation behavior, response synthesis, and missing-context behavior.
• Conduct adversarial testing for prompt injection, jailbreaks, instruction conflicts, sensitive-data leakage, unsafe outputs, and unauthorized tool use.

Must-have candidate profile

• Strong understanding of LLM behavior, prompt design, tokenization, context windows, RAG, embeddings, and model limitations.
• Hands-on experience with OpenAI APIs, Azure OpenAI, AWS Bedrock, Anthropic, LangChain, LlamaIndex, Semantic Kernel, Copilot Studio, or similar platforms.
• Ability to debug LLM outputs using structured testing, error analysis, and iterative refinement.
• Strong writing, analytical, communication, and stakeholder-management skills.
• Understanding of prompt-security risks including prompt injection, jailbreaks, data leakage, hallucination, and instruction conflicts.
• Ability to create repeatable prompt templates and evaluation evidence suitable for enterprise governance.

Preferred experience

• Background in NLP, conversational AI, UX writing, technical writing, product design, knowledge management, business analysis, or financial-services operations.
• Experience in financial services, legal, compliance, risk, operations, customer support, banker productivity, or enterprise knowledge domains.
• Familiarity with Microsoft Copilot Studio, Power Platform, prompt registries, A/B testing, human review workflows, and evaluation tooling.

Initial screening questions

• Show how you improved a weak LLM output through prompt design and testing.
• How do you evaluate prompt performance beyond subjective quality?
• How would you create reusable prompt templates for KYC, pitch book generation, Customer 360, or sanctions screening?
• How do you prevent prompt injection, data leakage, or instruction conflicts?
• How do you work with engineers to move prompts into production safely?
• How would you govern prompts used by citizen developers in Copilot Studio or Power Platform?