PROJECT_CHECKLIST.md

Module 2 Project Requirements Checklist

Modeling Approaches

• Naive baseline (random retrieval)
• Classical ML model (TF-IDF + cosine similarity)
• Deep learning model (sentence-transformers dense embeddings + FAISS)
• All three documented with locations in repo

Experimentation

• Focused experiment conducted (k-sensitivity analysis, k=1-10)
• Experiment is well-motivated and clearly described
• Results properly interpreted (in notebooks/analysis.md)

Interactive Application

• Publicly accessible via the internet (DEPLOYED)
• Runs model inference only (no training)
• Good UX — beyond basic Streamlit (assignment says "a basic Streamlit app is not going to cut it")
• Streamlit app exists (main.py) with game selector, retriever selector, example questions, eval metrics

Written Report

• Formal report (NeurIPS/ICML paper, white paper, or technical report)
  • Problem Statement
  • Data Sources
  • Related Work
  • Evaluation Strategy & Metrics (with justification)
  • Modeling Approach
  • Data Processing Pipeline (with rationale)
  • Hyperparameter Tuning Strategy
  • Models Evaluated (all 3 with rationale)
  • Results (quantitative comparison, visualizations, confusion matrices)
  • Error Analysis (5 specific mispredictions with root causes and mitigations)
  • Experiment Write-Up (plan, results, interpretation, recommendations)
  • Conclusions
  • Future Work
  • Commercial Viability Statement
  • Ethics Statement

In-Class Pitch (5 min)

• Presentation prepared
  • Problem & Motivation
  • Approach Overview
  • Live Demo
  • Results, Insights, Key Findings

Code & Deployment

• GitHub repository with full codebase
• Link to live deployed web application
• App runs inference on trained model
• Git best practices (branches used, PRs made)

Repository Structure

• README.md
• requirements.txt
• main.py (UI entry point)
• scripts/ directory with pipeline modules
• models/ directory
• data/ directory (raw, processed, outputs)
• notebooks/ directory
• .gitignore

Code Quality

• Jupyter notebooks only in notebooks/ (none used — all Python scripts)
• Code modularized into classes and functions
• No loose executable code outside functions/__main__
• Descriptive variable names, docstrings, comments
• External code / AI attribution at top of files (if applicable)

Novelty / Contribution

• Novel approach explained (multi-game RAG with 3-retriever comparison)
• Related work referenced (in project description)