Family E · Cross-cutting disciplines · Coming soon
Machine Learner Certification™
The distance between using AI and knowing how to build it.
Notice
This certification carries no regulatory recognition. It certifies knowledge and competence in applied machine learning and artificial intelligence; it does not replace any certification required by a regulator.
MLC
01 — ABSTRACTMachine Learner Certification™
Artificial intelligence has stopped being an innovation project in financial services and become infrastructure. The ten largest banks have put AI tools in the hands of more than 800,000 employees, and the sector's spending on the technology now exceeds USD 38 billion a year. What is scarce is not access to models. It is people who can build them, adapt them, run them in production and answer for them in front of a committee or a regulator.
The Machine Learner Certification™ attests to exactly that capability. Its body of knowledge spans fourteen modules across three levels: the mathematics and statistics beneath machine learning; classical and deep learning; language models and their adaptation to a domain; agentic systems; and, at the top, the engineering of scale, safety and production. It is written for the person who builds the model, not the person who calls an API.
The RiskMathics Certificate Institute certifies; Quantum Computing Labs prepares. The preparation program opens on November 3, 2026 with 452 hours across fourteen modules and a code-graded capstone in every module. The certification exam is a separate product: it can be taken with or without the program, and it is written to levels 1, 2 and 3 of Bloom's taxonomy. The certification carries no regulatory recognition today; it certifies competence.
MLC
02 — INDUSTRYWhat is happening in the industry
Financial-services spending on AI reached USD 38.36 billion in 2024 and is projected at USD 190.33 billion by 2030, a 30.6 % compound annual rate. Banking is the world's second-largest industry by AI investment. A single global bank employs more than 2,000 AI and data-science specialists, and 125,000 of its employees use its in-house language-model assistant every day.
Regulation is arriving at the same pace. The EU AI Act classifies credit scoring, fraud detection and anti-money-laundering profiling as high-risk, with full obligations for those systems from August 2026 (a postponement to December 2027 is under discussion) and penalties of up to 3 % of global turnover for non-compliance. In the United States, SR 11-7 remains the model-risk framework applied to machine learning in banking; in Latin America, Brazil and Mexico are advancing laws of their own.
Talent has not kept up. AI-skilled professionals command a 56 % wage premium, two thirds of business leaders say they would not hire someone without AI skills, and 73 % of financial-services leaders cite AI talent scarcity as a critical barrier. The gap between believing AI should be used (65 %) and actually using it (20 %) is, at bottom, a gap in verifiable training.
MLC
03 — WHYWhy this certification
The credentials on offer each solve part of the problem. AI-governance certificates attest to people who oversee models without asking for a line of code. Cloud-vendor certifications validate a platform with no financial context and no model risk. The reference quantitative program was designed before language models existed. Across a review of more than thirty credentials, the group's AI and finance blueprint found none that brings together, for financial services, machine learning, language models and the requirement to build and deploy: those that demand code lack financial context, and those with financial context do not demand code.
The Machine Learner Certification™ takes that position. It does not ask whether a candidate knows the vocabulary; it asks whether the candidate can solve the problem. That is why the preparation program closes every module with running code and a panel defense, and why the exam, independent of the program, is built from remember, understand and apply items over a published body of knowledge whose definitive weighting is fixed before the first sitting.
- Covers the full chain, from optimization and Bayesian inference to agents, scaling and production.
- Includes what a risk committee will review first: evaluations, interpretability, red-teaming and drift monitoring.
- Separates the user of an API from the engineer who can fine-tune, serve and control a model of their own.
- Assessed by an exam independent of the program; passing it is not a certificate of attendance.
- Designed for a field whose knowledge half-life the blueprint puts at 12 to 18 months; validity and the renewal scheme are published before the first sitting opens.
MLC
04 — OUTCOMESWhat it prepares for
- 01
Derive and implement a gradient-descent optimizer from scratch and demonstrate its convergence against the closed-form solution.
- 02
Build a complete tabular pipeline — data, features, tuned boosting, SHAP and calibration — with honest validation and a business reading of the top variables.
- 03
Estimate a causal effect from observational data, control for confounders and report the uncertainty to a committee.
- 04
Implement a neural network with hand-written backpropagation in NumPy, replicate it in PyTorch and justify every architectural decision.
- 05
Fine-tune an open-source language model to a domain with LoRA, evaluate it against a baseline and defend the cost, quality and control trade-off.
- 06
Build an agent with RAG and tools exposed through the Model Context Protocol that completes a real task end to end, with evaluation and guardrails.
- 07
Optimize a model's inference — quantization, distillation, serving — and measure latency, cost and quality before and after.
- 08
Audit a model with evaluations, red-teaming and interpretability, document its vulnerabilities, and deploy it with CI/CD, drift monitoring and a working rollback.
MLC
05 — INSTITUTIONSThe cut-off line
An institution that requires this certification changes the interview question. It no longer asks whether a candidate has used artificial intelligence — two thirds of employees at the ten largest banks already have AI tools at hand — but whether the candidate can build, validate and sustain a system that makes decisions with other people's money. That difference is paid for in model risk, in inference cost and in the time between a notebook and a service in production.
The credential also orders the internal market. It lets an institution distinguish, against one standard, the analyst who learned to ask a model for answers from the engineer who can explain why the model fails to converge, which data would improve it and what an auditor will look for once the system is classified as high-risk.
- A body of knowledge published by domain, with a weighting fixed and published before the first exam, comparable across candidates and across institutions.
- Evidence of the ability to put models into production, not only to train them.
- Safety, interpretability and monitoring competencies aligned with what the EU AI Act and SR 11-7 demand.
- An exam administered by a third party independent of both the teacher and the employer.
MLC
06 — PROFILEWho it is for
Who it is for
- Quants, traders, risk and portfolio managers, credit analysts, actuaries and treasurers whose work touches financial data.
- Data scientists and engineers, model and validation teams, and technology groups responsible for taking models to production.
- Engineers, mathematicians, physicists and economists with a quantitative base who want to become complete machine-learning practitioners.
- Heads of technology, data and transformation — and professionals in legal, compliance and healthcare — who need to understand the mechanism, not only the output.
Not for
- Anyone seeking an executive or conceptual overview of artificial intelligence without writing code.
- Anyone who needs only prompt literacy or fluency with commercial tools.
- Anyone seeking to certify quantum computing: that competence belongs to the Quantum Financial Professional (QFP™), currently in development.
MLC
07 — KNOWLEDGEBody of knowledge
Seven domains derived from the fourteen Core modules. Weights are indicative: they reflect each domain's share of the preparation program's 452 hours. The exam's weighting is to be confirmed and is published before the first sitting.
01Mathematical foundations and probabilistic reasoning≈ 16 %
- Linear algebra, SVD, projections and least squares
- Multivariable calculus, gradients and the chain rule
- Optimization: gradient descent, convexity, Lagrange and KKT
- Probability, entropy and KL divergence
- Bayesian inference, MCMC and uncertainty quantification
- Causal inference: DAGs, propensity scores and difference-in-differences
02Python, data and classical machine learning≈ 16 %
- Scientific Python, reproducible environments and version control
- Data engineering: ingestion, cleaning, features and pipelines
- Tokens, vectors and embeddings: how an LLM works inside
- Regression, classification, regularization and metrics
- Trees, ensembles and gradient boosting with hyperparameter tuning
- Unsupervised learning, dimensionality reduction and anomaly detection
- Robust validation and interpretability with SHAP
03Deep learning and multimodal models≈ 15 %
- Feedforward networks and backpropagation from scratch
- PyTorch: tensors, autograd and training loops
- Regularization, normalization and optimizers
- CNNs, transfer learning and recurrent networks
- Vision Transformers, detection and segmentation
- GANs, diffusion models and vision-language systems
- Document AI and the ethics of deepfakes
04Language: transformers, LLMs and domain adaptation≈ 15 %
- Tokenization, embeddings and the attention mechanism
- GPT and BERT architectures; designing an LLM
- Scaling, decoding and the context window
- Advanced prompt engineering and prompt evaluation
- Efficient fine-tuning: LoRA, QLoRA and PEFT
- Preference alignment: RLHF, DPO and reward models
- Serving, guardrails and the cost of a fine-tuned model
05Agentic AI, reasoning and reinforcement learning≈ 14 %
- Enterprise RAG: vector stores, chunking and evaluation
- Agents: tool use, planning, ReAct and memory
- Model Context Protocol and multi-agent systems
- Agents in production: cost, latency, observability and security
- MDPs, Q-learning, policy gradients and deep RL
- Reasoning models and test-time compute
06Scaling, data and production systems≈ 18 %
- Scaling laws, the economics of compute and mixed precision
- Distributed training: data, model and pipeline parallelism
- Inference optimization: quantization, distillation and KV-cache
- Data-centric AI: curation, synthetic data and evaluation by slice
- Lifecycle: versioning, feature stores and reproducibility
- Deployment with containers, CI/CD and continuous training
- LLMOps: prompts as code, continuous evaluations and guardrails
- Drift monitoring, model registry and governance
07Safety, interpretability and model evaluation≈ 6 %
- Alignment and specification: RLHF, DPO and Constitutional AI
- Mechanistic interpretability, probing and attribution
- Benchmarks, LLM-as-judge and evaluation contamination
- Red-teaming: jailbreaks, prompt injection and mitigations
MLC
08 — TWO PRODUCTSPreparation program · Certification exam
Preparation program
Preparation program
Program delivered by Quantum Computing Labs
Delivered by Quantum Computing Labs, the group's institute for artificial intelligence and quantum computing. It covers the entire exam syllabus in fourteen modules across three levels — Foundations, Deep & Generative AI and Frontier AI Engineering — through live classes, labs, a code-graded capstone in every module and mock exams built to the same recipe as the exam. Completing it is not a condition for sitting the exam.
Machine Learner Certification™ Core
452 h · 14 modules · 150 classesThe fourteen mandatory modules across three levels, from November 3, 2026 to December 7, 2027, closing with the Integrative Project and a Demo Day before an industry jury. This is the format that covers the exam's complete body of knowledge.
MLC Full Stack
788 h · 262 classesThe Core plus the first cohort's four specialization verticals — finance and markets, healthcare and life sciences, deep tech and quantum, and cybersecurity — taken in parallel between January and May 2028. The verticals extend application by industry; they are not part of the exam syllabus.
À la carte and level certificates
24 to 40 h per module · 136 to 172 h per levelAny Core module can be taken on its own at USD 100 per hour plus VAT, and each full level as a standalone certificate: AI Foundations (144 h), Generative AI Builder (136 h) and Frontier AI Engineering (172 h). Candidates preparing for the certification exam are advised to cover all three levels.
- 150 live sessions: Monday, Tuesday and Thursday from 7:00 to 10:00 pm, plus up to two in-person Saturdays a month from 9:00 am to 2:00 pm
- Recordings of every session, with 72-hour access for active students
- A code-graded capstone in every module, defended before an industry panel, in rotating teams
- Integrative Project and Demo Day before a jury
- Mock exams by domain built to Bloom levels 1, 2 and 3
- Weekly office hours, a buddy system and a pause policy of up to one module
November 3, 2026 to December 7, 2027 for the Core; verticals in parallel from January to May 2028. Breaks from December 11, 2026 to January 11, 2027, full Holy Week and Easter week, one week off every two months, and CNBV non-business days.
Certification exam
Certification exam
Administered by the RiskMathics Certificate Institute as a third party independent of the program's provider. It can be taken with or without the program. The Institute's master blueprint does not yet fix the length or structure of this exam; both are published before the first sitting opens.
If the candidate does not pass
The Certification Board is defining the failure and retake policy: how many attempts each window allows, the minimum wait between them, the fee for sitting the exam again and what happens to levels already passed. It is published here as soon as it is approved; in the meantime, the figures may be requested at growth@riskmathics.com. Exam governance →
Items are written to levels 1, 2 and 3 of Bloom's taxonomy: remember, understand and apply. Each domain contributes a fixed number of items per level, so that any two candidates face the same standard. Bloom's Taxonomy →
MLC
09 — MOCK EXAMSMock exams
The program's mock exams follow the exam's recipe: the same distribution by domain and the same three Bloom levels. They are not revision quizzes; they are rehearsals with calibrated items that let a participant see, domain by domain, where they remember, where they understand and where they cannot yet apply.
They are taken through the program and before the exam, with feedback by domain. Mock exams never reproduce live exam items; they replicate the exam's difficulty and structure.
- Level 1 · Remember: define, identify, list; for instance, recognize what a token is or what KL divergence measures.
- Level 2 · Understand: explain why a model fails to converge or what data leakage does to a validation.
- Level 3 · Apply: choose the right optimizer, fine-tuning technique or metric for a given case.
MLC
10 — PATHLearning path
- Phase 1
Foundations · Learn144 h · November 2026 to April 2027
Mathematics for ML, Python and AI fundamentals, probabilistic and causal ML, and classical ML. Four capstones, from an optimizer written from scratch to a production-grade tabular pipeline.
- Phase 2
Deep & Generative AI · Apply136 h · April to July 2027
Deep learning, multimodal and generative models, NLP and LLMs, and fine-tuning with RLHF and prompt engineering. Every module produces a production-grade capstone.
- Phase 3
Frontier AI Engineering · Create172 h · July to December 2027
Agents, RAG and MCP; reasoning and reinforcement learning; scaling and inference; data-centric AI; safety, interpretability and evaluations; MLOps and LLMOps.
- Phase 4
Integrative Project, Demo Day and mock examsAt the close of the Core · date to be confirmed
The production capstone feeds the Integrative Project, defended before a jury. Mock exams by domain close the preparation.
- Phase 5
Certification examDate to be confirmed
Administered by the RiskMathics Certificate Institute. The program's elective verticals continue in parallel from January to May 2028 for those who choose the Full Stack.
Investment
11Program and exam, priced separately
Fees in US dollars; VAT is applied according to invoicing. Program and exam are contracted separately; neither figure includes the other.
MLC
12 — FAQFrequently asked questions
Can the exam be taken without the program?
Yes. The exam is a separate product of the RiskMathics Certificate Institute. The Quantum Computing Labs program covers the full syllabus and includes mock exams, but it is not an admission requirement.
Who delivers the program and who certifies?
Quantum Computing Labs, the RiskMathics group's institute for artificial intelligence and quantum computing, delivers the preparation program. The RiskMathics Certificate Institute builds, audits and administers the exam as an independent third party.
What prior preparation is required?
Formal admission requirements are to be confirmed. The program is designed for quantitative profiles — engineering, mathematics, physics, economics, data — and also for people coming from business, legal or healthcare: Level 1 builds the mathematics, scientific Python and AI fundamentals from the ground up, and the syllabus marks the Python and fundamentals module as essential for anyone without a technical background.
Does it include quantum computing?
The exam's body of knowledge does not. The program offers an elective deep-tech and quantum vertical, and quantum competence is certified separately through the Quantum Financial Professional (QFP™), currently in development.
Can the program be taken in parts?
Yes. Any module can be taken à la carte at USD 100 per hour plus VAT, and each level as a standalone certificate: AI Foundations (144 h, USD 14,200 VAT included), Generative AI Builder (136 h, USD 13,400 VAT included) and Frontier AI Engineering (172 h, USD 16,950 VAT included). The full Core carries the best price per hour.
Does it carry regulatory recognition, and how long is it valid?
It carries no regulatory recognition: it certifies knowledge and competence. Validity and the renewal scheme are to be confirmed; the group's AI and finance blueprint contemplates continuing education in short cycles, since it puts the half-life of AI knowledge at 12 to 18 months.
Machine Learner Certification™
The distance between using AI and knowing how to build it.