PYQ solutions — four real papers, fully solved¶
T1-2026 (Jan) + T3-2025 (Sept) · both shifts · every question in original text, with the officially marked answer and the full reasoning.
The papers colour-code correct answers (green) — so every answer here is the official one, not our guess. The reasoning is ours. Code snippets that appear as images in the PDFs are reconstructed and marked (reconstructed).
How to read PYQs this term
The professor's ET-3 advice: use past papers to predict the core concepts, not the exact questions — this term's paper changes the question shapes, and about half of it is subjective. So treat these papers as a map of what gets asked about, and learn the concepts behind each question. The patterns below transfer; the exact questions may not. (His other rule: the most recent three papers are enough — these four cover that era.)
How to use these pages¶
- Cover the answer and attempt each question yourself — they're collapsible precisely so the page works as practice, not as reading.
- Open Answer to check. Only if you got it wrong (or slowly), open Why for the reasoning.
- The ⭐ questions repeat across papers and revision sessions — miss one twice and go read its topic sheet.
The four papers¶
| Paper | What it is | Special value |
|---|---|---|
| T1-2026 AN — solved | The most recent paper (Jan 2026, morning) — 28 questions including the first-ever short answers | The professor solved this live in ET-1 |
| T1-2026 FN — solved | Afternoon twin — ~60% shared topics, different angle on each | Links to AN for shared concepts; full treatment of the new ones |
| T3-2025 AN — solved | Sept 2025, morning — four scenario blocks (geospatial ⛔, weather API, LLM sentiment, RAG) | The RAG block is fully current — W4 is a whole RAG week |
| T3-2025 FN — solved | Sept 2025, afternoon — data-engineering practice + git/Docker/pandas scenarios | The professor solved this live in ET-2 |
The priority list — how often each topic is asked¶
Built from all four papers + both revision sessions. This table is the study order.
🔥🔥 Asked everywhere (guaranteed marks)¶
| Topic | Appears in | Where to learn it |
|---|---|---|
| CORS / same-origin | T1-AN ×2, T1-FN ×2, ET-1, course W2 | T1-AN Q13 |
| Docker: EXPOSE vs -p | T1-AN, T3-FN ×2, ET-2 | T1-AN Q28 |
| Docker: RUN pip install | T1-AN, T1-FN, T3-FN, ET-1 | T1-FN Q11 |
| Docker: parity/packaging | T1-FN, T3-FN, ET-1 | T1-FN Q5 |
| Caching: TTL, max-age, cache-buster | T1-AN ×2, T1-FN ×2, T3-AN, T3-FN, ET-1 + ET-2 | T1-AN Q19 |
| .env + .gitignore | T1-AN, T1-FN, T3-AN, T3-FN, ET-1 | T1-AN Q5 |
| HTTP status codes | T3-AN ×2, T3-FN, T1-FN, ET-2 | T3-AN Q4 |
| asyncio.gather = max | T1-AN, T1-FN, ET-1 | T1-AN Q21 |
| ML data leakage | T1-AN, T1-FN, ET-1 | T1-AN Q16 |
| Parquet (columnar) | T1-AN, T3-FN, ET-2 | T1-AN Q4 |
| System vs user prompts | T1-AN, T1-FN, ET-1, W3 | T1-AN Q23 |
| Prompt injection + filters | T1-AN, T1-FN ×2, ET-1, W7 | T1-FN Q12 |
🔥 Asked in 2–3 sources (high value)¶
Structured outputs · serverless limits (RAM + timeout) · race conditions · Pydantic/fail-fast validation · git lifecycle + branching + PR · pandas groupby/dropna · ETL lookback + idempotency · API gateway + redundancy · dependency pinning · reproducibility manifests · timezone handling · PII minimization · RAG fundamentals (chunking, retrieval, staleness)
Asked once (know it when you see it)¶
SQLite auto-create · YAML rules · shell redirection · localtunnel · Hypothesis invariants · DuckDB NULL handling · Pydantic coercion · CSV-for-stakeholders · rate-limit etiquette · lineage metadata · recommendation segmenting
⛔ Skip these in old papers¶
| Topic | Status | The 30-second version |
|---|---|---|
| OpenRefine | Professor: excluded this term | Interactive GUI data cleaning; its concepts (normalize→dedupe, replayable operations) are covered under pandas/ETL |
| Geospatial (Folium, GeoPandas, CRS) | Not in current course content | Degrees aren't metres; swapped coordinates give realistic garbage (T1-AN Q15) |
| Seaborn, spreadsheets | Old-syllabus leftovers | Recognize the names only |
The exam within the exam¶
Worth repeating from the exam pattern page:
- No negative marking — answer everything.
- The short answers (new in T1-2026) are LLM-graded — the grading guide teaches the format, and both T1 papers carry worked model answers (AN Q24–25, FN Q24–25).
- The FN/AN papers share ~60% of topics — studying one shift thoroughly covers most of the other.