IICTI Certified Artificial Intelligence and Machine Learning Engineer

Holders demonstrate that they can apply and evaluate distributed training and accelerator parallelism strategies, transformer internals and efficient attention architectures, lLM post-training, preference optimisation and PEFT and inference optimisation and high-throughput model serving, plus 7 further areas, in realistic professional scenarios, as defined in the AQSCERT scheme for Artificial Intelligence and Machine Learning Engineer.

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

1. 01Distributed training and accelerator parallelism strategies8% of the exam

   • Architect and tune large-scale training jobs that balance throughput, memory footprint, cost and fault tolerance across multi-node GPU clusters.
   • Scope: Data, tensor, pipeline and expert parallelism, FSDP/ZeRO sharding, DeepSpeed and Megatron-LM, BF16/FP8 mixed precision, activation checkpointing, NCCL collectives, fault-tolerant checkpointing and GPU utilisation (MFU) analysis.
2. 02Transformer internals and efficient attention architectures9% of the exam

   • Evaluate model architecture choices against quality, context-length, latency and memory requirements for a given workload.
   • Scope: Multi-head, grouped-query and multi-query attention, RoPE and long-context extension, FlashAttention, KV-cache mechanics, mixture-of-experts routing and load balancing, state-space and hybrid architectures.
3. 03LLM post-training, preference optimisation and PEFT8% of the exam

   • Design post-training pipelines that adapt foundation models to domain tasks while preserving general capability and alignment properties.
   • Scope: Supervised fine-tuning data curation, LoRA/QLoRA and adapter methods, RLHF with PPO, DPO and GRPO, reward modelling, distillation, synthetic data pipelines and mitigation of catastrophic forgetting and reward hacking.
4. 04Inference optimisation and high-throughput model serving8% of the exam

   • Architect serving stacks that meet latency, throughput and cost targets under variable production traffic.
   • Scope: Quantisation (GPTQ, AWQ, FP8, INT4), speculative decoding, continuous batching, PagedAttention, vLLM, TensorRT-LLM, SGLang and Triton Inference Server, prefill/decode disaggregation, autoscaling and latency SLOs (TTFT, TPOT).
5. 05Retrieval-augmented generation and agentic system architecture7% of the exam

   • Design RAG and agentic systems with measurable grounding, controlled autonomy and predictable failure behaviour.
   • Scope: Embedding models, hybrid lexical/vector search, HNSW and IVF-PQ indexes, rerankers, chunking and context engineering, tool calling, Model Context Protocol, multi-agent orchestration, memory and guardrail design.
6. 06Evaluation science for ML and LLM systems17% of the exam

   • Evaluate model and system quality with statistically defensible methods that predict real-world performance.
   • Scope: Offline metric design, calibration, benchmark contamination, LLM-as-judge bias and agreement, human evaluation protocols, online A/B and interleaving experiments, statistical power and significance, evaluation harnesses.
7. 07Feature platforms and training-serving consistency14% of the exam

   • Design feature infrastructure that guarantees training-serving parity and prevents leakage at production scale.
   • Scope: Feature stores (Feast, Tecton, Databricks), point-in-time correct joins, target leakage, streaming feature computation, online/offline skew detection, feature reuse and lineage.
8. 08MLOps and LLMOps: continuous training, release and reproducibility8% of the exam

   • Lead the design of an ML delivery platform that makes model releases reproducible, auditable and safely reversible.
   • Scope: CI/CD/CT pipelines, MLflow and Kubeflow, model and prompt registries, data and artefact versioning, lineage, shadow, canary and champion-challenger deployment, automated rollback and reproducible environments.
9. 09Production monitoring, drift and failure diagnosis7% of the exam

   • Diagnose production model degradation and define monitoring, alerting and retraining triggers proportionate to business impact.
   • Scope: Covariate, label and concept drift, PSI, KS and MMD tests, embedding drift, delayed-label strategies, hallucination and quality telemetry, OpenTelemetry GenAI conventions, root-cause analysis of degraded models.
10. 10Adversarial machine learning and AI system security6% of the exam

    • Evaluate AI systems against adversarial threat models and architect proportionate technical mitigations.
    • Scope: OWASP Top 10 for LLM Applications 2025, MITRE ATLAS, direct and indirect prompt injection, data and model poisoning, model extraction, membership inference, model supply-chain integrity and layered defences.
11. 11Responsible AI engineering and regulatory conformity8% of the exam

    • Design engineering controls and evidence that demonstrate fairness, transparency and regulatory conformity of ML systems.
    • Scope: Fairness metrics and mitigation, SHAP and counterfactual explainability, differential privacy, model and system cards, EU AI Act technical documentation (Annex IV), ISO/IEC 42001 and NIST AI RMF engineering controls.

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