Foundation Models · Advanced
Foundation Model Engineering
Transformer architecture, pretraining, adaptation, grounding, evaluation, inference and serving — the complete engineering picture of modern foundation models.
- Level
- Advanced
- Mathematics depth
- Deep
- Engineering depth
- Research-grade
- Modality
- Cohort-based · intensive
- Duration
- Announced per cohort
- Status
- Open for inquiry
Audience
Who this is for
- Engineers who will adapt, evaluate and serve foundation models
- ML engineers moving from classical ML to LLM systems
- Technical leads making architecture decisions on model stacks
Career paths
Prerequisites
- — Solid Python and PyTorch
- — Working linear algebra
- — Prior ML/DL coursework or experience
Outcomes
Skills acquired
- Explain every major component of a foundation model stack from first principles
- Adapt models with fine-tuning and PEFT under evaluation discipline
- Design serving architecture with cost and latency engineering
Tools used
Curriculum Architecture
Module progression
Expandable, visual, ordered. Each module is a prerequisite-aware step, not an isolated video.
- 01Transformer Foundations
- 02Tokenization & Representation
- 03Attention Systems
- 04Transformer Architecture
- 05Pretraining
- 06Instruction Tuning
- 07Fine-Tuning
- 08PEFT
- 09Grounding & Context
- 10Evaluation
- 11Inference
- 12Serving
- 13Agents
- 14Multimodal Systems
- 15Production Architecture
Projects
- A fine-tuned adapter with regression evaluation
- A grounded answer system with citation quality metrics
- A serving plan with tiered routing and cost model
Assessment philosophy
Assessment is engineering review: written error analyses, measured system behavior, defended design decisions. We evaluate whether you can explain and justify what you built — because production will.
Stack position: Foundation Models.
FAQ
Frequently asked questions
Do I need a mathematics background?
It depends on the program. Foundation-tier programs start from the mathematics itself; advanced tiers list working linear algebra as a prerequisite. The Mathematics for AI program exists precisely to close that gap.
Is this a bootcamp?
No. CAS Studies is an engineering institute. Programs are built around architectures, derivations and projects with written error analysis — not tutorial replays.
How long does a program take?
The two diploma tracks run on fixed lengths — AI Architectural Engineering spans 18 months, AI Application Engineering spans 1 year. All other program durations are announced per cohort, by modality.
Will I build real systems?
Yes. Every program ends in projects that resemble production work: evaluated models, grounded answer systems, supervised agents — with measurement, not vibes.