Final Project — Guidelines¶

Weight: 6 of the 25-mark Sessional component (institutional scheme: Sessional 25 · Mid Exam 35 · Final Exam 40 — see assessment-plan.md) · Teams: 2–3 students · CLOs: 1, 2, 3

1. What you must build¶

An end-to-end data science project, in order:

Problem → Data acquisition → Cleaning → EDA → Visualization →
Basic ML → Evaluation → Streamlit app → AI-assisted analysis
(optional: LLM / agent / n8n) → Documentation → GitHub

Required minimum components (all 13 mandatory)¶

Problem definition — a specific, answerable question with a stated audience and decision.
Dataset / data source — documented, openly licensed (no paid APIs).
Data acquisition — reproducible loading (script/function, cached).
Data cleaning — dtypes, missing values, duplicates, with a logged reason per decision.
EDA — univariate + group comparisons + correlations.
Visualization — at least 3 labeled figures answering specific questions.
Basic ML model — at least one scikit-learn model (regression, classification, or clustering) with a train/test split.
Model evaluation — honest metric(s), a baseline comparison, and a one-paragraph interpretation.
Interactive Streamlit application — widgets that filter/explore the data and show your figures/tables (see ../assignments/assignment-02-api-streamlit-app/streamlit-tutorial.md).
GitHub repository — commit history showing the project evolving.
README — runnable-by-a-stranger documentation (template provided).
Final report — ~1,000 words incl. ethics + reproducibility reflection (template provided).
Presentation — 6 minutes + 2–3 minutes Q&A (template provided).

Optional advanced components (pick 0+; at least one recommended for CLO-3)¶

PandasAI natural-language queries (audited and logged)
Ollama local model assisting an analysis step
LLM prompt/verification protocol applied to your own work
A simple AI agent loop over your data tools
n8n (or equivalent) automation for data collection/processing

None are required — the mandatory AI-assisted requirement is the AI-assisted analysis step (an LLM/PandasAI/Ollama-assisted step with disclosure + verification), which can be as small as one documented, verified query.

2. Teams¶

2–3 students. All members must be able to explain any part of the project (viva, §7).
Roles are encouraged but every member contributes code: e.g., data/EDA, modeling, app, documentation — with commit history as evidence.
Peer evaluation forms are submitted at presentation time (peer-evaluation-form.md) and inform the individual component of the grade.

3. Scope limits (what "achievable" means)¶

No paid APIs, no API keys, no cloud services required. Use keyless APIs (e.g., Open-Meteo), public datasets, or the course built-ins.
No advanced frameworks: pandas, NumPy, Matplotlib, Seaborn, scikit-learn, Streamlit, and (optionally) requests, PandasAI, Ollama, n8n are the allowed stack. No PyTorch/TensorFlow/Django/Flask.
Datasets: ≤ 10 MB committed; larger datasets load via a script with a documented source. License must permit reuse (CC0/CC BY preferred).
Everything must run offline after the first fetch (cache your data).

4. Milestones & deadlines¶

Milestone	Due	Gate
Kickoff, teams, draft question	W12 S23 (in class)	Lab 23 workbook
Proposal	W13 S26	graded gate (5% of project)
Reproducible repo layout + seeds	W12 S24 (workshop)	Lab 24
Mid-project check-in (cleaning + EDA done)	W14 S28 (in class)	informal
Presentation	W16 S31–32	graded
Final submission (report + repo)	W16 S32	graded

Proposal is a gate: no proposal, no project grade — see rubric.

5. Submission¶

Private GitHub repository: data-science-final-project (or projects/final/ inside your private course repo — pick one, tell your instructor). Invite the instructor.
At least 10 commits spread over the working period.
Final submission bundle: repo (at deadline commit) + report + README + executed notebooks + app runnable via streamlit run app.py.
Late policy: −10%/day per ../assessment-plan.md.

6. Academic integrity & AI use¶

Course policy (../assessment-plan.md) applies in full:

Disclose every AI-assisted step (tool, prompt, how used) in the README and report.
Verify AI-generated code runs and you can explain it; findings must be checked against the data (the verification protocol from Session 25).
Plagiarism of peers/external code is subject to institutional policy.
You may be asked to explain any line of your work in the viva; being unable to explain AI-generated code is treated as unverified use.

7. Viva¶

After the presentation, any team member may be asked 2–4 questions from viva-questions.md (own-component questions for each member). The viva is worth 10% of the project grade and is individual.