Amazon ML Summer School 2026: A Past-Paper Archive, Not a Course

The problem it solves: no official question bank exists

Amazon runs the ML Summer School selection test through Unstop, and the README describes the format plainly: Part A is 20 MCQs on machine learning basics, probability, statistics and linear algebra, Part B is 2 programming questions, and the whole thing lasts 60 minutes. Both sections have separate cutoffs. Nothing in the material suggests Amazon publishes a question bank afterwards. Candidates finish the test and the questions disappear.

This repository exists to fill that gap. It gathers recalled questions from 2023, 2024 and 2025, plus an official sample test that the README attributes to Scaler. It is aimed at students graduating in 2027 or 2028 from an Indian institute, which is the eligibility window the README states. If you are outside that window, or you are looking for a structured machine learning course, this is the wrong artefact. It is a revision pack for one specific assessment.

What is actually in the repository

The contents list is short and specific. Programming.md holds coding questions from the 2023 selection assessment. SummerSchool2023.md holds 2023 MCQs. MLSummerSchool2024.pdf is described as the complete question pool and exam layout from the 2024 cohort. Amazon_ML_Summer_School_Sample_Test.pdf is the official sample shared by Scaler, marked as solved. Two further PDFs, AMSS_2025_Exam_Questions.pdf and AMSS_2025_Questions.pdf, cover the morning and afternoon shifts of the 2025 assessment.

The module content sits in the AMSS 2025 directory. The README lists eight modules with dates: Supervised Learning on 9 August 2025, Deep Neural Networks on 10 August, Dimensionality Reduction on 16 August, Unsupervised Learning on 17 August, Probabilistic Graphical Models on 23 August, Sequential Learning on 24 August, Causal Inference on 30 August, and Reinforcement Learning on 31 August. Each module links to a Markdown file named MODULE1.md through MODULE8.md. The README states that practice tasks and coding exercises appear at the bottom of each module summary, which is the part that turns reading into preparation.

One structural detail is worth flagging. The repository is named for 2026, but the substantial content is labelled 2025 and earlier. The README says the repository hosts materials for the 2025 programme while the registration banners and links point at 2026. That mismatch is not a defect exactly, since past papers are the point, but it means the 2026-specific content is thin: a registration link, an application link, and a release tagged v1.0.0 described as an SOP guide and template.

How the module summaries are meant to be used

The intended data flow is linear. You open a module file, read the notes, and work the tasks appended at the end. The README frames this as read, learn, revise, practice. There is no build step, no test runner, no dependency file, and no code that executes. Everything is Markdown and PDF.

That has a practical consequence for how you should treat the folder. The notes are one person's summaries of sessions delivered by Amazon scientists, not the scientists' own slides. Where a summary compresses a topic like Probabilistic Graphical Models or Causal Inference into a single Markdown file, the compression is the risk. These are the two modules where a summarised treatment is most likely to lose the derivation steps that the MCQs actually test. The README offers no indication of how long each summary is or how closely it tracks the delivered session.

A second practical issue is PDF rendering. The README anticipates this and tells you that GitHub sometimes fails to preview PDFs on certain browsers, and that the fix is to click Download in the top-right corner of the preview screen and read the file locally. Take that note at face value: if a PDF link shows an error, the file is probably fine and the viewer is not.

Getting it onto your machine

There is nothing to install. The repository has no package manifest and no build configuration in the supplied material, so the only setup is cloning it and opening the files. The README does not give a clone command, but the standard form against the default branch works:

git clone https://github.com/cu-sanjay/Amazon-ML-Summer-School-2026.git cd Amazon-ML-Summer-School-2026

From there, the entry points named in the README are Programming.md, SummerSchool2023.md, MLSummerSchool2024.pdf, Amazon_ML_Summer_School_Sample_Test.pdf, AMSS_2025_Exam_Questions.pdf and AMSS_2025_Questions.pdf, with the module files under the AMSS 2025 directory. Note the space in that directory name. In a shell you will need to quote it, for example ls "AMSS 2025", or escape the space.

The README also links outward to a separate repository, cu-sanjay/Amazon-ML-Summer-School-2025, for the More ML Resources section covering roadmaps and additional notes. Those links point at the 2025 repository rather than this one, which is consistent with the pattern noted above: the 2026 repository reuses 2025 paths. If a link 404s, try the same path under the 2025 repository name before assuming the content is gone.

The verification problem, and the licence

The hardest limitation is provenance. The README describes the 2025 question files as questions from the assessment held during the morning shift and the afternoon shift. It does not say how they were collected, whether they are complete, or whether the answers are official. Recalled exam questions are reconstructions. Wording drifts, options get reordered, and a transcribed answer can be wrong. Nothing in the material lets you check any of it against an Amazon source. The one file with a stated official origin is the Scaler sample test.

There is a second problem with the questions themselves. The README states that Amazon strictly checks for plagiarism in coding submissions and warns against sharing code during the assessment. That warning is about the live test, but it sits awkwardly next to a public archive of past coding questions. Copying a solution from Programming.md into a live submission is exactly the behaviour the warning targets, and the repository offers no guidance on where the line is. Treat the coding questions as practice prompts to solve yourself, not as answers to reproduce.

The licence field reads NOASSERTION. That means GitHub could not map the repository to a recognised licence, so there is no stated grant of rights over the notes, PDFs or question transcriptions. Third-party material, including the Scaler sample test and any Amazon branding, is not the maintainer's to license. For personal revision this is unlikely to matter. For republishing the content, bundling it into a paid course, or mirroring the PDFs on another site, the permission is simply not established. That is a description of the licence field, not legal advice.

What it is not, and what to use instead

This is not a machine learning course. The eight module files summarise sessions, and the README points to a separate resources folder for deeper material, but the repository's own centre of gravity is assessment preparation. If you want to actually learn causal inference or reinforcement learning, a summarised Markdown file dated to a single August session is a poor primary source.

For the learning side, the honest alternative is a structured course with exercises that are graded or at least self-checking: Andrew Ng's Machine Learning Specialisation on Coursera, or the fast.ai course, both of which build the same foundations (supervised learning, neural networks, dimensionality reduction) across weeks rather than single sessions. The difference in approach is that those courses sequence material and test you as you go, while this repository assumes you already have the foundations and need to recall them under a 60-minute clock. A second alternative for the test format itself is the official Scaler sample test, which the README identifies as the only officially sourced assessment file here. If you only have time for one thing, that PDF is the highest-confidence document in the repository.

Where the repository genuinely has no substitute is the recalled question pool. There is no other place in the supplied material where 2023, 2024 and 2025 questions sit together with a module-by-module revision path attached.

Maintenance cost and what to check before relying on it

The repository was last pushed on 2 July 2026 and carries a single release, v1.0.0 from 15 June 2026, described as an SOP guide and template. There is no stated update cadence, no contributing guide in the material, and no issue triage process described. A repository of this shape depends on one maintainer collecting questions after each cycle, which means the 2026 question files may or may not appear after the 2026 test runs. Do not plan around them.

The upgrade cost is effectively zero, because there is nothing to upgrade. You clone it, read it, and the only thing that changes over time is whether new question files land. The maintenance risk is staleness rather than breakage: a link that points at the 2025 repository, a module file that has not been revised since August 2025, a PDF that a browser refuses to preview.

Before you build a study plan on it, do three checks. Open the AMSS 2025 directory and confirm that MODULE1.md through MODULE8.md exist and render, since the README's links to them are the repository's main claim. Open Amazon_ML_Summer_School_Sample_Test.pdf and confirm it downloads, because that is the one officially sourced file. Then read the 2025 question PDFs with a textbook open beside you and mark every answer you cannot independently justify. The repository is a useful index of what has been asked. It is not a source of truth about what is correct.

Editorial conclusion

Use this repository if you are preparing for the Amazon ML Summer School selection test and want a single place to read recalled MCQs, coding questions and eight module summaries. Do not treat it as an authoritative syllabus: the question files are reconstructions, the repository carries a NOASSERTION licence, and the README still points several content links at the 2025 repository. Before you rely on it, open the AMSS 2025 folder and confirm that MODULE1.md through MODULE8.md actually render, then cross-check any answer you plan to memorise against a textbook or lecture source.

Official sources

1. cu-sanjay/Amazon-ML-Summer-School-2026 on GitHub
2. Issues
3. Project website
4. README
5. Releases