IICTI Certified Generative AI Engineer

Holders demonstrate that they can apply and evaluate transformer and LLM architecture internals, retrieval-augmented generation architecture, agentic systems and tool orchestration and model adaptation: fine-tuning and preference optimisation, plus 7 further areas, in realistic professional scenarios, as defined in the AQSCERT scheme for Generative AI Engineer.

Who it is for: Professionals working, or preparing to work, in artificial intelligence roles.

1. 01Transformer and LLM architecture internals9% of the exam

   • Evaluate model architectures and their trade-offs to select the right foundation model for a given workload.
   • Scope: Attention variants (MHA, GQA, MLA), RoPE and long-context extension (YaRN), mixture-of-experts routing, KV cache behaviour, tokenisation effects, scaling laws, reasoning models and test-time compute.
2. 02Retrieval-augmented generation architecture14% of the exam

   • Architect RAG systems that maximise grounded answer quality while controlling latency and cost.
   • Scope: Chunking and contextual retrieval, embedding model selection, hybrid lexical-dense search, cross-encoder reranking, ANN index types (HNSW, IVF-PQ), query rewriting, GraphRAG, citation grounding and freshness.
3. 03Agentic systems and tool orchestration12% of the exam

   • Design agentic workflows with bounded autonomy, reliable tool use and recoverable failure modes.
   • Scope: Tool and function calling, Model Context Protocol (MCP) and agent-to-agent protocols, ReAct and plan-and-execute patterns, multi-agent orchestration (LangGraph), memory and state management, human-in-the-loop checkpoints, failure recovery.
4. 04Model adaptation: fine-tuning and preference optimisation9% of the exam

   • Evaluate when and how to adapt a model and design training pipelines that deliver measurable gains.
   • Scope: SFT, LoRA/QLoRA/DoRA and other PEFT methods, RLHF, DPO, ORPO and GRPO, synthetic data generation, knowledge distillation, catastrophic forgetting, decision criteria for prompting vs RAG vs fine-tuning.
5. 05Context engineering and structured generation at scale13% of the exam

   • Design context and prompt systems that produce reliable, schema-conformant outputs across model versions.
   • Scope: Structured outputs via JSON Schema and constrained decoding, prompt templating and versioning, context-window budgeting and compaction, prompt caching, dynamic few-shot selection, programmatic prompt optimisation (DSPy).
6. 06LLM evaluation and quality engineering12% of the exam

   • Design evaluation frameworks that detect quality regressions and justify model or prompt changes with evidence.
   • Scope: Golden datasets and eval harnesses, LLM-as-judge calibration and bias, pairwise and rubric scoring, RAG metrics (faithfulness, context precision), regression gating in CI, online A/B testing, annotation protocols, benchmark contamination.
7. 07Inference serving optimisation and cost engineering8% of the exam

   • Diagnose and optimise LLM serving stacks to meet latency, throughput and unit-cost targets.
   • Scope: vLLM/SGLang PagedAttention and continuous batching, speculative decoding, quantisation (GPTQ, AWQ, FP8), tensor and pipeline parallelism, TTFT/TPOT latency targets, model routing and cascades, GPU capacity planning.
8. 08LLM application security and adversarial robustness8% of the exam

   • Evaluate GenAI systems against adversarial threats and design layered defences proportionate to risk.
   • Scope: OWASP Top 10 for LLM Applications 2025, direct and indirect prompt injection, excessive agency, sensitive data leakage, guardrail models (Llama Guard, NeMo Guardrails), MITRE ATLAS, red-teaming governance and reporting.
9. 09Multimodal generative systems4% of the exam

   • Architect multimodal pipelines that combine text, image, audio and document modalities reliably in production.
   • Scope: Vision-language models, diffusion and flow-matching image generation, speech recognition and synthesis, document understanding pipelines, content provenance (C2PA) and watermarking.
10. 10LLMOps, observability and lifecycle management6% of the exam

    • Lead the operational lifecycle of GenAI applications from release through monitoring, incident handling and model migration.
    • Scope: Tracing with OpenTelemetry GenAI semantic conventions and Langfuse, prompt and model registries, AI gateways with rate limits and fallbacks, canary model rollouts, drift detection, user feedback loops, model deprecation planning.
11. 11AI governance, risk and regulatory compliance5% of the exam

    • Evaluate GenAI solutions against regulatory and organisational requirements and design proportionate governance controls.
    • Scope: EU AI Act risk tiers and GPAI obligations, ISO/IEC 42001 AI management systems, NIST AI RMF and Generative AI Profile (NIST AI 600-1), model and system cards, data provenance and copyright, GDPR, AI impact assessments.

Question types

Single choice 100%

Delivery

Online (remote proctored), Test centre, On-site (corporate)

Retakes

Up to three attempts within any 12-month period. A 14-day waiting period applies after a first unsuccessful attempt and 30 days after a second. Each retake requires a new booking and fee. Retakes draw a different randomised form from the question bank.

Provisional cut score — psychometric validation is required before live delivery. Blueprint v1.0.

1. 1

   Register

   Create a free account and start the programme.

2. 2

   Practice

   Online practice test (coming soon).

3. 3

   Training

   Optional

   Online course or classroom — or skip it.

4. 4

   Certification

   Independent exam and certification decision.

Training is optional and does not by itself confer certification. Find a training partner