Ai Engineer

Job Title – Ai EngineerSalary – AED 25,000 - 28,500 Per MonthContract Length – 12 Months ContractLocation – DubaiStart Date - ASAPMy client is a global data, AI, digital transformation, and consulting company that helps organisations use data and technology to improve business performance, customer experience, and decision-making.Role Overview: The AI Engineer is the primary builder within the AWA AICoE. You will design, build, and test the LangGraph agent graphs, LLM extraction pipelines, prompt management workflows, and MCP tool integrations that power AWA's use cases — starting with Intelligent Document Processing (IDP) for cheque clearing and expanding across the bank's operations. You will work directly with the dual-layer orchestration stack (Orkes Conductor + LangGraph) and Azure AI Foundry to deliver production-grade agentic systems.Key ResponsibilitiesAgent graph development: Design and implement LangGraph agent graphs for Supervisor-Worker orchestration, self-reflection, actor-critic, and HITL interrupt patterns using TypedDict state schemas and conditional edge routing.LLM extraction pipeline: Build and maintain LLM-based entity extraction pipelines using Azure AI Foundry, implementing structured output enforcement, per-field confidence scoring, and Pydantic output validation against Use Case Manifest schemas.Prompt engineering and governance: Author, version, and govern prompts through the AWA Prompt Management System (PMS) using semantic versioning; run prompt sensitivity and correctness testing using the AWA IDP Test Strategy.MCP tool integration: Develop and maintain the five core MCP tool servers (query_structured, retrieve_precedents, query_documents, checkpoint_state, observe) and extend them for new use case requirements.Model evaluation and testing: Execute Band 1 and Band 2 testing (Model Testing, Prompt Testing, Agent Testing, AI Security Testing, Adversarial Testing) per the AWA IDP Test Strategy; maintain the Golden Dataset and Regression Dataset.HITL framework: Implement HITL interrupt() calls at mandatory and dynamic review points in LangGraph graphs; configure Orkes Wait tasks; validate HITL state persistence and checkpoint-based resumption.AI guardrail integration: Integrate Azure AI Content Safety, PII detection and tokenisation, prompt injection shielding, and output sanitisation into agent pipelines at the AWA AI Gateway layer.Cross-field validation rules: Implement business validation rules (amount-words-vs-figures, date validity, MICR format, referential integrity) within LangGraph validation agent nodes.Required Skills and ExperienceTechnical — Essential5+ years Python development — production-grade, not just scriptingHands-on experience with LangGraph, LangChain, or equivalent agentic AI frameworksPractical LLM prompt engineering: structured outputs, few-shot design, output schema enforcementFamiliarity with Azure AI Foundry / OpenAI API or equivalent LLM API (Anthropic, Cohere)Understanding of agent design patterns: ReAct, self-reflection, tool use, HITLREST API development and consumption; JSON schema designGit version control; CI/CD pipeline awarenessTechnical — AdvantageousExperience with Orkes Conductor, Apache Airflow, or Temporal for workflow orchestrationAzure AI Document Intelligence or equivalent OCR platform experienceComputer vision model integration (object detection APIs)Knowledge of RAG architectures: chunking, vector indexing, Azure AI SearchAzure platform familiarity: AKS, ADLS Gen2, Azure ML, Azure MonitorPydantic, FastAPI, TypeScript/Node.jsDomain and Soft SkillsGenuine intellectual curiosity about AI systems and their failure modesAbility to explain AI behaviour to non-technical Ops and business stakeholdersStructured problem-solving and debugging approach for non-deterministic AI systemsBanking domain interest — willingness to deeply understand the business processes being automatedThanks,Ollie