AI Video Comment Analyzer

AI-powered YouTube comment analysis with ML sentiment detection, topic modeling, and AI-generated summaries.

Architecture

Features

• Comment Extraction: Fetch up to a configurable comment limit per video using yt-dlp (default 100)
• Sentiment Analysis: BERT-powered multilingual sentiment classification (positive/negative/neutral/suggestion)
• Topic Modeling: BERTopic clustering to identify key discussion themes
• AI Summaries: Local LLM-powered summaries via Ollama (llama3.2:3b)
• Multi-Page Dashboard: Dedicated pages for Overview, Charts, Topics, and Comments
• Real-time Progress: SSE streaming with ML metrics during analysis

Home

Overview

Charts

Suggestions

Quick Start

Prerequisites

• Node.js 20+
• pnpm
• Python 3.11+
• uv (Python package manager)
• Ollama (optional, for AI summaries)
• NVIDIA GPU with CUDA (optional, for faster ML inference)

Frontend Setup

# Install dependencies
pnpm install

# Run development server
pnpm dev

Open http://localhost:3000 in your browser.

Backend Setup

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install dependencies (includes ML models: torch, transformers, bertopic)
uv sync

# Run API server
uv run uvicorn api.main:app --reload --port 8000

API available at http://localhost:8000

Ollama Setup (Optional)

For AI-generated summaries of comment sentiment:

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull the model
ollama pull llama3.2:3b

# Start Ollama server (if not running)
ollama serve

Architecture

Multi-Page Routes

/                           - Home (URL input + history sidebar)
/analysis/[id]              - Overview page (At a Glance + Summary Cards)
/analysis/[id]/charts       - Charts page (2x2 grid with context)
/analysis/[id]/topics       - Topics page (list + detail panel)
/analysis/[id]/comments     - Comments page (filters + sorting)

Tech Stack

• Frontend: Next.js 15, React 19, TypeScript, Tailwind CSS v4, shadcn/ui, Recharts
• Backend: FastAPI, Python 3.11+, yt-dlp, SQLAlchemy
• Database: SQLite
• AI/ML:
  • nlptown/bert-base-multilingual-uncased-sentiment (sentiment)
  • BERTopic with all-MiniLM-L6-v2 embeddings (topics)
  • Ollama with llama3.2:3b (AI summaries)

Design System

• Typography: Fraunces (display), DM Sans (body), JetBrains Mono (code)
• Colors: Navy (#1E3A5F) + Terracotta (#D4714E) accent, warm cream background
• Sentiment: Green (positive), Red (negative), Blue (suggestion), Gray (neutral)

Project Structure

src/
├── app/
│   ├── page.tsx                    # Home (input + history)
│   ├── layout.tsx                  # Root layout with fonts
│   ├── globals.css                 # Design system
│   └── analysis/[id]/
│       ├── layout.tsx              # Analysis layout (nav + tabs)
│       ├── page.tsx                # Overview page
│       ├── charts/page.tsx         # Charts page
│       ├── topics/page.tsx         # Topics page
│       └── comments/page.tsx       # Comments page
├── components/
│   ├── navigation/                 # GlobalNav, AnalysisTabs
│   ├── blocks/                     # EvidenceStrip, SummaryCard, SentimentFilter
│   ├── charts/                     # SentimentPie, EngagementBar, etc.
│   └── results/                    # TopicRanking, CommentCard, etc.
├── context/
│   └── analysis-context.tsx        # Shared analysis state
├── hooks/
│   ├── useAnalysis.ts              # Analysis + ML metrics
│   └── useAnalysisData.ts          # Data fetching
└── types/
    └── index.ts                    # TypeScript interfaces

api/
├── main.py                         # FastAPI app
├── config.py                       # Environment config
├── routers/
│   └── analysis.py                 # SSE streaming endpoints
├── services/
│   ├── youtube.py                  # yt-dlp extraction
│   ├── sentiment.py                # BERT sentiment
│   ├── topics.py                   # BERTopic modeling
│   └── summarizer.py               # Ollama summaries
└── db/
    └── models.py                   # SQLAlchemy models

API Endpoints

Method	Endpoint	Description
POST	/api/analysis/analyze	Start analysis (SSE stream)
GET	/api/analysis/result/{id}	Get analysis results
GET	/api/analysis/result/{id}/comments	Get comments for analysis
GET	/api/analysis/history	List past analyses
DELETE	/api/analysis/history/{id}	Delete an analysis
GET	/api/analysis/search	Search YouTube videos

Environment Variables

Copy .env.example to .env and configure:

# YouTube
YOUTUBE_MAX_COMMENTS=100

# Ollama (AI Summaries)
OLLAMA_URL=http://localhost:11434
OLLAMA_MODEL=llama3.2:3b
OLLAMA_ENABLED=true

# ML Processing
SENTIMENT_BATCH_SIZE=32
MAX_TOPICS=10

Development

Running the App

# Terminal 1 - Frontend
pnpm dev

# Terminal 2 - Backend
uv run uvicorn api.main:app --reload --port 8000

# Terminal 3 - Ollama (optional)
ollama serve

Running Tests

# Run all tests
uv run pytest tests/ -v

# Run with coverage
uv run pytest tests/ -v --cov=api --cov-report=term-missing

Code Quality

# Lint
uv run ruff check api/ tests/

# Format
uv run ruff format api/ tests/

# Frontend lint
pnpm lint

CI Pipeline

GitHub Actions runs on every push/PR:

• Lint: Ruff check + format verification
• Test: pytest with 65% coverage threshold

License

MIT