Where models meet
market structure
Practical machine learning education built around cryptocurrency data. Not theory for its own sake - techniques that actually touch real signals, real noise, and the decisions that follow.
The people who built this and still teach it
Assthree was founded in 2020 by practitioners who had already spent years working with financial time-series data and predictive modelling. The curriculum is not assembled from textbooks - it reflects decisions, failures, and refinements that came from applying these methods to actual cryptocurrency datasets.
Every module is delivered by someone who has written the code, debugged the pipeline, and interpreted results under uncertainty. That background shapes what gets taught and, more importantly, what gets left out.
Dariusz Kowalczyk
Lead Instructor - ML Systems
Eight years building classification and regression models for financial data. Focuses on the gap between prototype performance and production reliability.
Priya Naidoo
Curriculum Designer - Data Pipelines
Designed the assessment architecture and hands-on quiz system. Specialises in turning complex preprocessing decisions into teachable, testable concepts.
What stays useful after the last quiz
Most short courses teach you to run a model. The harder problem is knowing when your model is lying to you - when overfitting is disguising itself as accuracy, or when a feature that worked in one market regime stops working in another.
The programme is structured so that the reasoning process transfers. The specific libraries will change. The habit of interrogating your own outputs is what carries forward into whatever data environment you encounter next.
Build a working mental model
Understand why an algorithm behaves the way it does on noisy financial data, not just how to call the function.
Develop evaluation discipline
Learn to design tests that expose weakness before deployment - walk-forward validation, out-of-sample checks, regime sensitivity.
Apply across different datasets
Assignments use multiple asset pairs and time periods deliberately - so the method generalises rather than memorises one market.
The conditions under which this works
This is not a programme that works for everyone, and it is worth being direct about that. The content assumes certain things about how you are willing to engage with it.
You are comfortable with Python basics
Not expert-level - but you should be able to read a loop, understand what a function returns, and debug a simple error message without losing an hour. The course teaches ML concepts, not Python from scratch.
You have time to work through the assignments
The quizzes and graded exercises are designed to take time. Skipping them and reading the explanations produces a different outcome than working through the problem first. Expect six to eight hours per module.
You are curious about how the models fail
Participants who get the most out of this programme tend to be more interested in edge cases and failure modes than in headline accuracy numbers. That orientation matters more than prior experience.
Finishing the programme - what that position looks like
Completion is not a credential in the traditional sense. What changes is the range of problems you can approach with reasonable confidence - and the vocabulary you have for explaining your approach to others.
Participants who finish the full curriculum have worked through datasets covering multiple asset classes, built and broken several model architectures, and written up their reasoning in structured assignments that mirror the documentation expectations of technical teams.
Design feature sets for OHLCV data and explain why each feature was included or excluded
Select between model families based on data characteristics - not just default to gradient boosting
Articulate the limits of a model's predictions in a way that non-technical colleagues can act on
Identify data leakage in a pipeline before it produces misleading results in production
Programme scope
6 modules
Spanning data ingestion through to model monitoring - each module builds on the last without assuming the previous one was easy.
A completion certificate is issued for participants who pass all graded assignments. The certificate reflects demonstrated work, not attendance.
What the investment actually covers
Pricing is structured around access to the full curriculum, not individual modules. The numbers below give a general sense of scope - the contact page has current figures and any active enrolment periods.
Duration
14 wks
Self-paced within a structured timeline. Deadlines exist but are generous enough to accommodate full-time work.
Weekly load
6–8 hrs
Realistic estimate based on participants who completed all assignments without prior ML experience.
Cohort size
Under 30
Kept deliberately small so that feedback on assignments is specific rather than automated and generic.
Access after end
12 months
All materials, quiz attempts, and assignment submissions remain accessible for review after the cohort closes.