Project Requirements Document¶
¶
Section 3.1: Project Overview & Requirements Recap (Finalizing the Menu and Diner Experience)¶
Before laying out the technical infrastructure, we must ensure absolute clarity on what we’re building and for whom. This section consolidates the decisions made in Chapter 1 and presents them as formal project documents.
3.1.1 Presenting the Finalized Product Requirement Document (PRD) The PRD serves as the single source of truth for what the “Trending Now” application will do. It details the project’s purpose, features, users, and success criteria.
Product Requirement Document: “Trending Now” Movies/TV Shows App
Introduction & Purpose:
To create an application that provides users with up-to-date information on newly released movies and TV shows, featuring LLM-generated genre classifications, review summaries, vibe-based scores, and relevant tags.
To serve as a comprehensive, educational MLOps project within this study guide.
Target Audience:
End-Users: Movie/TV show enthusiasts looking for new content across multiple OTT platforms.
Learners (of this guide): Individuals seeking to understand practical MLOps implementation.
Core Features:
Data Ingestion: Regularly scrape/fetch new movie/TV show releases and user reviews from specified sources.
Genre Classification (Educational Model): Train an XGBoost/BERT model to classify content genre based on plot/reviews. (This is primarily for demonstrating MLOps training pipelines).
LLM-Powered Content Enrichment (Production Inference Path):
Generate concise summaries of aggregated user reviews.
Generate a “vibe score” (1-10) based on review sentiment and content.
Generate descriptive “vibe tags” for intuitive content discovery.
(Production Path) Classify genre using an LLM.
Homepage Visualization: Interactive D3.js bubble chart displaying movies/shows.
Bubbles sized by LLM-generated score.
Default view: All recent shows, potentially loosely clustered by overall rating.
Interactive Buttons:
Group by OTT platform.
Group by Genre (primary LLM-generated genre).
Group by Vibe Tags (most prominent tags).
Re-cluster by Score buckets.
Hover Interaction: Popup card on bubble hover showing title, primary genre, score, OTT platform.
Detail Page: Dedicated page per movie/TV show displaying:
Title, poster, plot summary (scraped).
LLM-generated genre(s).
LLM-generated review summary.
LLM-generated vibe score.
LLM-generated vibe tags.
Links to source reviews.
Success Metrics (from Chapter 1 Project Section):
App User Engagement (Conceptual for Guide): (e.g., DAU, Session Duration)
Genre Accuracy (LLM Path - User Perception): High user satisfaction with assigned genres.
Review Summary Quality: High user satisfaction with clarity and conciseness.
Vibe Score & Tag Relevance: High user satisfaction and utility for discovery.
Educational XGBoost/BERT Model Metrics: Macro F1 > X%, Precision/Recall per genre > Y%.
MLOps System Metrics: Pipeline reliability, data freshness, monitoring effectiveness.
Non-Goals (for this phase of the project):
User accounts and personalization (beyond basic vibe search).
Real-time streaming of new reviews (batch ingestion is sufficient).
Complex recommendation algorithms (focus is on presentation of LLM-enriched data).
Perfect, production-grade scraping (best effort for educational purposes).
3.1.2 Presenting the App/User Flow Diagrams Visualizing how users will navigate the application helps solidify requirements and identify potential UX issues early.
“Trending Now” App User Flow
This flow diagram outlines the primary interactions, focusing on data discovery through the bubble chart and accessing detailed, LLM-enriched information.