Most Nepali master’s theses are written the same painful way: one enormous Word document, PDFs scattered across Downloads and Viber forwards, citations assembled by hand at 2am before submission, and a literature “review” that is really a literature pile. Students I teach across M.Ed, MPA and MBA programs lose weeks to this — not to the research, to the mess. This post builds the alternative toolkit: Obsidian for thinking, Zotero for citing, AI for structuring — with the integrity guardrails that keep your degree safe. For course materials in these programs, see all course notes.

1. The system in one picture

• Obsidian (free): your thinking space — one note per source, linked into theme maps. This is where the literature review gets written, almost as a side effect.
• Zotero (free): your reference library — PDFs stored once, metadata clean, citations inserted in Word/Google Docs in your department’s style.
• AI assistant: your structuring helper — outlines, counterarguments, statistical plan checks. Never your author, never your data analyst of record.

Rule zero: tools serve the argument. A beautiful vault with no thesis question is procrastination with plugins.

2. Obsidian: literature review as a Map of Content

Setup that survives a full thesis:

1. One vault per thesis, folders minimal: sources/, themes/, drafts/, admin/ (proposal, supervisor comments, defense prep).
2. One note per source with a fixed template: full citation, 3-sentence argument in your own words, key quote with page number, your critique (what it misses, what it assumes), and links to 2–3 related notes.
3. Maps of Content (MOCs): one index note per theme or chapter section linking the relevant source notes with a sentence each on how they connect. The MOC for “digital payment adoption factors” with twelve linked, summarized sources is your literature review’s skeleton — expanding it into prose is weeks faster than starting from a pile.
4. Daily research log: date, what you read, what changed in your thinking, next question. Supervisors love this; it also becomes your methodology chapter’s honesty record.

Resist plugin-collecting. Core plugins plus templates cover thesis work; every hour configuring graphs is an hour not reading.

3. Zotero: citations that assemble themselves

Set up before collecting a single PDF:

• Install Zotero + browser connector + the Word/Google Docs plugin on day one.
• Fix metadata on import (Nepali author names and institutional reports import messily — correct them immediately, not in submission week).
• File PDFs as attachments to items, never loose in folders; use collections per chapter.
• Learn two operations deeply: Add/Edit Citation while drafting and Add/Edit Bibliography at the end. Switching citation styles afterward is one dropdown, not a rewrite.
• Group library shared with your supervisor turns feedback rounds from “fix reference 47” emails into direct comments.

4. Ethical AI: structure, never substance

Where AI genuinely helps thesis work:

• Turning your MOC into a chapter outline with gaps flagged (“no source yet on rural sample”)
• Brainstorming counterarguments to your hypothesis before the defense does
• Checking your statistical plan: appropriate tests for your data type and sample, assumptions to verify (see which-test tool for MPA 509 students)
• Tightening your own drafted paragraphs for clarity

Where it must never go: generating findings, fabricating quotes or citations (AI-invented references are the fastest route to a failed defense — verify every suggested citation exists in Zotero before citing), analyzing raw data unsupervised, or writing passages you submit as your own without disclosure. Log AI assistance and disclose per your supervisor’s and department’s norms; norms differ across TU/KU programs, so ask explicitly rather than assuming.

5. Statistics without tears (MPA/MBA focus)

Most quantitative theses stall at the same points: choosing tests, justifying sample size, and presenting output honestly.

• Map each research question to exactly one test before collecting data; the mapping belongs in the proposal, not the appendix.
• Learn assumption checks (normality, expected cell counts) as part of the test, not as decoration — examiners ask about these.
• Present effect sizes and confidence intervals alongside p-values; a “significant” trivial effect is honest reporting, not a finding.
• Keep syntax/output files versioned with dates. “Final_final_v2” is how results become irreproducible.

What to do this week

1. Install Obsidian + Zotero, create the vault and library, and migrate your five most important sources into the one-note-per-source template.
2. Draft one MOC for your weakest chapter section — the gaps you find are your reading list.
3. Ask your supervisor the AI-disclosure question explicitly, in writing, before using AI on a single paragraph.

Related reading: all course notes, MPA 509 statistics, which-test tool, AI marketing guide, and contact for training.