Arcadia AI — Master's Program
EN
Program  /  For the University

The program from the university's perspective

Administrative framework: credits, learning outcomes, admission requirements, regulations, and partnership model — for the rectorate, academic office, and admissions committee

Legal launch model

The program is launched in partnership with an existing university. The working group is considering three options; no decision has been made — the framework below is for discussion.

1A new master's specialization within the partner university's license. The shortest track: it uses the existing license, and the program undergoes special accreditation before instruction begins.
2A joint / dual program with a foreign university. Provides an international diploma and a strong marketing asset; requires an inter-university agreement and alignment of two regulatory frameworks.
3A separate educational organization. Maximum autonomy, but the longest and most capital-intensive path; under current regulatory conditions, treated as a long-term scenario.

Program credit map — 120 ECTS (draft)

Working workload allocation: 30 ECTS per semester, 120 for the program — a benchmark compatible with the Bologna master's framework. A draft for reconciliation with state standards and the partner university's regulations.

S1Linear Algebra — 5 · Probability Theory and Statistics — 5 · Algorithms — 5 · Classic ML/DL — 6 · Elective — 4 · Practicum 1 — 5  =  30
S2Optimization — 5 · NLP/LLM/Agents — 6 · MLOps — 5 · Reliable ML — 4 · Elective — 4 · Practicum 2 — 6  =  30
S3Multimodal — 5 · AI Research — 6 · ML Engineering — 5 · Applied AI — 4 · Elective — 4 · Practicum 3 — 6  =  30
S4Research Seminar — 2 · AI Safety — 4 · Capstone — 6 · Thesis Seminar — 2 · Pre-thesis practicum — 4 · Thesis — 12  =  30

Program learning outcomes

A summary framework compiled from course-level outcomes (full wording on the individual course pages). A graduate:

Builds and trains modern models — from classical ML to transformers, LLMs, and agentic systems — and justifies architecture choices
Takes a model to production: data pipelines, deployment, monitoring, and operation under SLA and quality constraints
Conducts reproducible research: from replicating a scientific paper to an original research contribution
Evaluates and ensures the reliability, safety, and fair assessment of AI systems
Works with high-performance infrastructure: GPUs, distributed training, LLM serving, compression
Defends a Master's thesis — an original contribution on either the research or production track

Admission requirements (draft)

Bachelor's degree; a STEM background is preferred but not required if entrance exams are passed
Mathematics at a technical bachelor's level: linear algebra, calculus, basics of probability theory
Confident programming in Python
English proficiency sufficient to read scientific literature and documentation
Format of entrance exams (exam / portfolio / interview) — to be determined by the working group

Faculty staffing — an honest status

The core team built the Yandex School of Data Analysis (YSDA), Yandex.Textbook, and competitive-programming training programs. As of today, mentors are confirmed and under discussion for 5 of the 21 core courses; recruitment is ongoing for the rest — each course's status is shown on its own page. Target staffing rule: every course is covered by a lecturer-practitioner plus a seminar instructor, and thesis supervision is a separate role, not combined with lecturing duties by default. Regulatory requirements on the academic degrees of master's-program faculty are treated as a mandatory condition for the final roster and are being closed jointly with the partner university.

Master's thesis regulations (draft)

1Topic selection and assignment of a thesis advisor — by the end of semester 2; an individual schedule for each master's student
2Research Seminar and Thesis Seminar in semester 4 — structured support for the writing process
3Pre-thesis practicum — final experiment and results collection
4Pre-defense with an external reviewer → defense before the committee
!The defense procedure is fixed according to the regulator's requirements in force at the time of defense (the certification system is currently being reformed) — the regulations are reconciled with the partner university before launch

Assessment and quality control

Each course concludes with a measurable assessment tied to its stated outcomes: an exam, a project, or a combination of both. Practicums and the capstone are graded against criteria announced in advance (reproducibility, functionality, documentation). Course materials go through a two-stage review process: generation → expert review by the course instructor; until reviewed, materials are explicitly marked as a draft. The program is designed for external curriculum review before submission for accreditation.

Partnership model: who brings what

Program team: curriculum and materials, mentor network (including engineers from Google Zurich), a learning platform with practicums and auto-grading, marketing and admissions, operational support
Partner university: license and accreditation track, diploma, faculty with academic degrees meeting regulatory requirements, classrooms and administrative infrastructure, representation before the regulator

Open questions for the working group

1Legal launch model: a specialization under the university's license / a joint program / a separate organization?
2Target tuition cost and size of the first cohort
3Filling teaching positions for 16 courses, including semesters 3–4
4License-holding university and status of negotiations
5A realistic launch horizon given the accreditation track
This is a draft for the working group: all parameters marked "draft" / "to be determined" are working assumptions for discussion, not approved decisions.