AI Workflow Engineering Architect

AI Workflow Engineering Architect On-site | Abu Dhabi AppliedAI, founded in 2021, is a pioneering AI technology company headquartered in Abu Dhabi, UAE. We are committed to innovation and excellence in artificial intelligence solutions across regulated industries such as healthcare, insurance, government, and financial services. Our flagship platform, Opus, automates and supervises mission-critical, document-heavy processes with embedded governance, auditability, and human oversight. We help enterprises achieve measurable productivity gains while increasing operational reliability and trust. Position Overview: The AI Workflow Engineering Architect is the senior technical authority for the design and optimization of production workflows within Opus, with particular responsibility for ML-intensive and computationally demanding business processes. The role establishes how Opus workflows should be engineered to achieve the required accuracy and reliability at the lowest practical inference cost and latency, using the capabilities available within Opus and supporting systems. You’ll lead the technical development of Workflow Engineers through hands-on design work, architecture reviews, training, reference implementations and systematic analysis of production performance. As the custodian of workflow engineering practice within AppliedAI, you’ll be responsible for converting advances in ML and computer science, and lessons from production, into standards, methods and reusable engineering patterns across the team. Key responsibilities: Workflow architecture and optimization: • Lead the decomposition of business processes into efficient Opus workflow graphs, defining execution boundaries, dependencies, state, parallelism and failure paths. • Formulate important workflow-design decisions as constrained optimization problems across accuracy, latency, throughput, inference cost and operational reliability, and establish the appropriate operating point rather than optimizing any metric in isolation. Model architecture, selection and routing: • Establish the technical methods by which Workflow Engineers select and compose models within Opus. • Design and review heterogeneous execution strategies including deterministic computation, specialist models, retrieval, model cascades, conditional routing, early exits and human review. • Require model choice and routing decisions to be supported by measured error rates, calibrated confidence, workload characteristics and marginal inference economics. Evaluation and experimental methodology: • Define the engineering standards by which Opus workflows are demonstrated to work. • Establish representative test sets, business-weighted loss functions, component and end-to-end benchmarks, ablation methods, confidence intervals and regression thresholds. • Ensure evaluation covers distribution shift, rare cases, correlated failures and high-cost errors rather than relying on aggregate model accuracy or successful demonstration cases. Inference and systems performance: • Lead technical analysis of workflow execution cost and performance, including model inference, context construction, retrieval, serialization, network calls, concurrency and orchestration overhead. • Establish methods for profiling and improving caching, batching, parallel execution, model size, quantization, context length and accelerator utilization. • Require performance to be characterized under realistic concurrency using throughput, cost per successful execution and p50/p95/p99 latency. Reliability and execution semantics: • Establish best practice for reliable Opus workflow execution, including typed interfaces, explicit state transitions, idempotency, checkpointing, bounded retries, timeouts, backpressure, compensation and partial-failure recovery. • Apply techniques such as property-based testing, fault injection, deterministic replay and execution-trace analysis where appropriate. • Ensure Workflow Engineers understand failure behaviour as an architectural property rather than something addressed after deployment. Production measurement and continuous optimization: • Define how workflow performance is measured after deployment and how evidence from production feeds back into engineering decisions. • Establish instrumentation connecting workflow versions, model and configuration choices, execution traces, errors, latency, inference consumption and business outcomes. • Lead diagnosis of regressions and determine whether corrective action belongs in workflow structure, data, model choice, routing, context, software implementation or capacity. Technical leadership and engineering capability: • Develop Workflow Engineers into strong ML and computer-science practitioners capable of making these decisions independently. • Lead difficult workflow designs, architecture reviews and technical post-mortems; maintain reference implementations, engineering standards and reusable Opus patterns; and train the team in optimization, evaluation, inference engineering and production ML. • Act as the final technical custodian of workflow engineering best practice, progressively converting expert knowledge into a repeatable engineering discipline across Opus. Required Skills: • Senior-level applied ML and software engineering background, with hands-on experience designing and operating production LLM/agentic systems (not research-only or prototype-only) • Strong grounding in ML evaluation methodology and systems performance (latency, cost, concurrency) as engineering disciplines, not afterthoughts • Distributed systems fundamentals: reliability, failure handling, and state management in production pipelines • Demonstrated experience mentoring engineers and setting technical standards, not just doing the work solo Desirable Skills: • Experience with multi-model orchestration (routing, cascades, fallback across providers like Anthropic, OpenAI, Gemini) • Background in a regulated industry (healthcare, finance, insurance, BPO) • Experience building engineering standards or a technical discipline from scratch in an early-stage environment • Published writing, research, or patents on ML systems or inference optimization What We Offer: • Opportunity to shape the marketing analytics strategy at a leading AI company. • Exposure to cutting-edge AI/ML applications in real-world business contexts. • Collaborative, innovative, and growth-driven work environment. • Competitive compensation, benefits, and career advancement opportunities.

Senior-level applied ML and software engineering background, with hands-on experience designing and operating production LLM/agentic systems (not research-only or prototype-only); strong grounding in ML evaluation methodology and systems performance (latency, cost, concurrency) as engineering disciplines, not afterthoughts; distributed systems fundamentals: reliability, failure handling, and state management in production pipelines; demonstrated experience mentoring engineers and setting technical standards, not just doing the work solo

Workflow architecture and optimization: Lead the decomposition of business processes into efficient Opus workflow graphs, defining execution boundaries, dependencies, state, parallelism and failure paths; formulate important workflow-design decisions as constrained optimization problems across accuracy, latency, throughput, inference cost and operational reliability, and establish the appropriate operating point rather than optimizing any metric in isolation. Model architecture, selection and routing: Establish the technical methods by which Workflow Engineers select and compose models within Opus; design and review heterogeneous execution strategies including deterministic computation, specialist models, retrieval, model cascades, conditional routing, early exits and human review; require model choice and routing decisions to be supported by measured error rates, calibrated confidence, workload characteristics and marginal inference economics. Evaluation and experimental methodology: Define the engineering standards by which Opus workflows are demonstrated to work; establish representative test sets, business-weighted loss functions, component and end-to-end benchmarks, ablation methods, confidence intervals and regression thresholds; ensure evaluation covers distribution shift, rare cases, correlated failures and high-cost errors rather than relying on aggregate model accuracy or successful demonstration cases. Inference and systems performance: Lead technical analysis of workflow execution cost and performance, including model inference, context construction, retrieval, serialization, network calls, concurrency and orchestration overhead; establish methods for profiling and improving caching, batching, parallel execution, model size, quantization, context length and accelerator utilization; require performance to be characterized under realistic concurrency using throughput, cost per successful execution and p50/p95/p99 latency. Reliability and execution semantics: Establish best practice for reliable Opus workflow execution, including typed interfaces, explicit state transitions, idempotency, checkpointing, bounded retries, timeouts, backpressure, compensation and partial-failure recovery; apply techniques such as property-based testing, fault injection, deterministic replay and execution-trace analysis where appropriate; ensure Workflow Engineers understand failure behaviour as an architectural property rather than something addressed after deployment. Production measurement and continuous optimization: Define how workflow performance is measured after deployment and how evidence from production feeds back into engineering decisions; establish instrumentation connecting workflow versions, model and configuration choices, execution traces, errors, latency, inference consumption and business outcomes; lead diagnosis of regressions and determine whether corrective action belongs in workflow structure, data, model choice, routing, context, software implementation or capacity. Technical leadership and engineering capability: Develop Workflow Engineers into strong ML and computer-science practitioners capable of making these decisions independently; lead difficult workflow designs, architecture reviews and technical post-mortems; maintain reference implementations, engineering standards and reusable Opus patterns; and train the team in optimization, evaluation, inference engineering and production ML; act as the final technical custodian of workflow engineering best practice, progressively converting expert knowledge into a repeatable engineering discipline across Opus.