Senior AI Engineer

Role - Senior AI Engineer About the Role GHX is building an LLM-powered document understanding platform — classification, structured data extraction, and intelligent orchestration at scale. This is a broad AI engineering role where you will own the full lifecycle from problem framing through to production quality. You will start where the highest leverage is: designing prompts that behave like specifications and building the evaluation infrastructure that tells us whether the system actually works. Over time, the scope expands into agent orchestration, system architecture, and the migration of our existing rule-based pipeline to LLM-based equivalents. Solid software engineering experience is a pre-requisite. You will design and build production services, own architectural decisions, and write code that others maintain. AI fluency amplifies good engineering – it does not replace it. This role requires 8+ years of software engineering experience as a foundation. The expectation is that you bring the same rigor to system design and code quality as you do to the prompt design and model evaluation. You know what should happen before evaluating what did happen — and the discipline to close that gap rigorously. How the Role Evolves Now Prompt engineering LLM evaluation Design classification and extraction prompts, build ground truth datasets, create evaluation pipelines, and establish the quality baseline the team will build on. 6–12 mo Agent orchestration pipeline design Extend into multi-agent document processing pipelines, MCP-based tool integration, and orchestration of parallel AI workstreams. 12 mo+ Platform architecture migration ownership Own the architecture of the document intelligence layer, lead the migration from rule-based systems, and define engineering standards for the team. Core Responsibilities Prompt Engineering Design classification and extraction prompts for diverse document types Write prompts that function as formal specifications — unambiguous and edge-case-aware Build and iterate few-shot, chain-of-thought, and structured output templates Own the prompt library, versioning, and rollback strategy LLM Output Evaluation Define and curate ground truth datasets for classification and extraction tasks Build automated evaluation pipelines tracking precision, recall, and field-level accuracy Validate outputs against intent — catch what is technically correct but conceptually wrong AI Agent Orchestration Design multi-agent pipelines for document processing workflows Integrate MCP servers and tool-use patterns for external service access Software Engineering Build production-grade APIs and services around LLM capabilities Apply Clean Architecture or equivalent — design for testability and maintainability Write well-structured Python; own technical debt decisions consciously Contribute to CI/CD, observability, and deployment pipelines Stakeholder Interface Serve as technical contact between business teams and the AI engineering layer Communicate system constraints in product language; translate product needs into system boundaries Surface quality metrics and model behaviour to non-technical stakeholders Skills Experience Required Prompt engineering (advanced) LLM evaluation frameworks Ground truth dataset design LLM APIs — OpenAI, Anthropic, Azure AI 8+ years overall software engineering experience preferably in Python AWS AI agent orchestration Clean Architecture / DDD REST API design Git, CI/CD, containerisation Nice to Have Document understanding / OCR tooling like Textract Experience working with AWS services (EC2, SQS, S3, Lambda functions, ECS) LangChain / LlamaIndex NLP / text classification What Good Looks Like Prompt as specification. Treats prompts like specifications — precise, contract-like, with edge cases considered upfront rather than patched later. Verification instinct. Reads AI output critically. Catches the subtly wrong answer before it becomes ... (truncated, view full listing at source)