Role Overview
NeuralCraft AI is building the next generation of enterprise AI products and we need a Prompt Engineer & LLM Systems Designer who can sit at the intersection of language, logic, and product. You won't be training models; you'll be crafting the instruction layers, reasoning pipelines, and evaluation systems that make our AI reliably useful in high-stakes domains like legal, finance, and healthcare.
What You Will Do
• Design and iterate on prompt pipelines for multi-step reasoning tasks across legal, finance, and healthcare verticals.
• Build and maintain RAG (Retrieval-Augmented Generation) systems using LangChain, LlamaIndex, and vector databases such as Pinecone or Weaviate.
• Architect prompt chains for agentic LLM workflows involving tool use, memory, and conditional branching.
• Define and run structured evaluation frameworks to measure output quality, hallucination rates, and task completion.
• Collaborate with ML engineers on fine-tuning strategies for domain-specific model adaptation.
• Document prompt templates, versioning strategies, and evaluation results for cross-team reproducibility.
• Stay current with emerging models (GPT-4o, Claude 3.x, Gemini, Mistral) and benchmark their performance for internal use cases.
What We Are Looking For
Must Have
• Demonstrated experience designing production-grade prompts not just chatbot experiments.
• Hands-on knowledge of LangChain or LlamaIndex for building LLM-powered pipelines.
• Familiarity with embedding models and vector search (Pinecone, Weaviate, Chroma, or equivalent).
• Understanding of LLM internals: tokenization, context windows, temperature, top-p sampling.
• Experience running structured evaluations automated and human-in-the-loop.
• Strong written communication skills; prompts are documents too.
Good to Have
• Experience with fine-tuning open-source models (LLaMA, Mistral) using LoRA / QLoRA.
• Working knowledge of Python for scripting evaluation pipelines and data wrangling.
• Exposure to compliance-sensitive AI applications (GDPR, HIPAA-adjacent use cases).
• Published prompt engineering work, open-source contributions, or a public GitHub portfolio.