I design production prompt architectures that enforce deterministic, structured, and reliable AI outputs — built for high-stakes contexts where hallucination and inconsistency are not acceptable. What's included:

System prompt architecture design with schema enforcement JSON output schema definition and validation layer Hallucination reduction strategies (few-shot anchoring, constraint injection, fallback logic) Evaluation framework for your specific use case Written documentation with reasoning for every design decision

Past work: Designed LAZARUS_SYSTEM_PROMPT encoding AHA-compliant clinical detection criteria with deterministic JSON schema output for real-time medical triage classification at sub-500ms latency. Built BCG GenAI RAG pipeline with NLP fine-tuning. Works with: Claude API, OpenAI GPT-4o, Gemini, Mistral, or any LLM endpoint. Delivered in 3–7 days. Most straightforward engagements need only one revision round.