WorkspaceREADME
Masters Thesis Workspace — Operating Systems & AI
A guide and working scaffold for producing a monograph-style MSc thesis in Computer Engineering over a 6–12 month single research cycle.
The guide/ assumes you have already chosen a topic and gathered papers, and it carries no
reading list — a stale bibliography would be worse than none.
If you have not settled on a topic yet, start in topic-selection/. That directory is the
one exception to the no-bibliography rule, and it is explicitly dated for that reason: its
crowdedness verdicts and citations were verified in August 2026 and will decay quickly.
What is here
| Path | What it is |
|---|---|
topic-selection/ | Candidate topics with verified crowdedness evidence and a scoped reading list. Read before guide/01-scoping.md. Dated — re-verify before committing. |
guide/ | Nine documents, read in order. The reasoning behind every choice you are about to make. |
templates/ | Fill-in worksheets you use weekly. Copy them, don't edit the originals. |
thesis/ | LaTeX monograph skeleton with the chapter structure already stubbed out. |
experiments/ | Directory layout, environment-capture scripts, and result conventions that keep your numbers reproducible. |
reader/ | Next.js portal for reading the markdown. Local: cd reader && npm run dev. Deploy: root Dockerfile or Compose on Dokploy; container port 3000, no host bind. |
Read the guide in this order
guide/00-the-map.md— the 6–12 month timeline, its phases, and the gate you must pass to leave each oneguide/01-scoping.md— narrowing your angle, writing a falsifiable thesis statement, establishing the gapguide/02-reading-system.md— turning a pile of papers into a related-work chapter that argues somethingguide/03-research-design.md— deciding what evidence would convince a skeptic, before you gather anyguide/04-thesis-structure.md— chapter by chapter, with page budgetsguide/05-writing-process.md— drafting order, daily cadence, and how to use your advisor wellguide/06-evaluation-playbook.md— measurement rigor for systems work; the longest document here, deliberatelyguide/07-defense-and-submission.md— logistics and the questions you will actually be askedguide/08-failure-modes.md— how theses in this area fail, and the early warning signs
Do these five things this week
Everything else can wait. These five compound, and each one gets more expensive the longer you delay it.
1. Find your university's formatting and submission rules and read them. Not in month eight. Now. Margins, required front matter, embargo policy, the exact submission deadline, how many committee members must sign, whether a plagiarism check is mandatory, and what the policy on AI-assisted writing is. Put every hard date in a calendar with a two-week-early reminder. People lose weeks to this, and the weeks they lose are the ones they least have to spare.
2. Write a first draft of your thesis statement. One sentence, falsifiable. Use
templates/thesis-statement-worksheet.md. It will be wrong and you will rewrite it
several times, which is exactly why you write it now rather than after the experiments.
3. Decide your angle. guide/01-scoping.md walks you through AI-for-OS versus
OS-for-AI versus a measurement study. These three have different risk profiles and
require different infrastructure in month one, so leaving the choice open is expensive.
If the topic itself is still open, work through topic-selection/ first — and note its
central finding, which is that no area at this intersection is currently uncrowded, so every
viable topic is framed as an incumbent's stated limitation rather than an unexplored problem.
4. Set up citation management before you have 40 papers. Zotero plus Better BibTeX,
auto-exporting to thesis/refs.bib. Retrofitting this later is miserable manual work.
5. Book a recurring meeting with your advisor. Weekly or biweekly, in the calendar,
recurring. Use templates/weekly-advisor-update.md so each meeting starts from a
written artifact rather than from you trying to remember what happened.
The one idea to hold onto
A thesis is not a report on what you did for a year. It is an argument that a specific, falsifiable claim is true, supported by evidence a skeptical expert would accept.
Every decision in this guide follows from that. If a chapter, experiment, figure, or paragraph does not help establish your claim or honestly bound it, it does not belong in the document — however much work it took you.