Artificial Intelligence Expert

Role Overview:

Own the design, testing, and productionization of prompt systems that power:

• AI search visibility for client brands (ChatGPT, Perplexity, Gemini, Copilot)
• Creative & analytics workflows (ad copy, briefs, summaries, reports)
• Internal agents/automations (RAG, structured outputs, guardrails, evals)
  

Key Responsibilities (KRAs)

• Prompt Architecture:

  System/few-shot patterns, tool/function calling, multi-turn flows, structured JSON outputs.

• Quality & Evals:

  Golden datasets, offline/online evals, regression tests; track accuracy, coverage, drift, latency, token cost.

• RAG & Contexting:

  Chunking/embeddings, retrieval, caching/fallbacks, context-window tuning.

• Guardrails & Safety:

  Schema validation, moderation/PII checks, safety policies.

• Cost & Performance:

  Optimize tokens, latency, hit rates; set budgets/alerts; observability & prompt versioning.

• Shipping:

  Turn prototypes into lightweight services/APIs; write runbooks; enable SEO/Creative/Reporting teams with reusable libraries.

Requirements (Must-Haves first, then Nice-to-Haves)

• Must-Haves

• 14 years in LLM apps / data / backend

  or

  a strong portfolio of shipped LLM projects.
• Hands-on with a major model provider (OpenAI/Anthropic/Google) incl. JSON/function calling.
• Practical RAG: embeddings + chunking + retrieval with a vector DB (any).
• Built an evaluation setup (golden set / A/B) and used it to improve quality.
• Structured outputs with schema validation; can debug formatting failures robustly.
• Comfortable with Python or JS, APIs/Postman, Git, basic SQL.
• Clear written communication and documentation
  

• Nice-to-Haves

• SEO/SEM or content ops exposure (schemas/FAQs, citations, ad creatives).
• Analytics/GA4 reporting or data-to-text generation.
• Guardrails: moderation, PII redaction, jailbreak resistance.
• LangChain/LlamaIndex, Pinecone/Weaviate/pgvector; caching; worker queues.
• Multi-lingual prompts; eval at scale; feature flags/canary releases.