neo-notebook

AI Security Intelligence Engine

A dual-mode AI security intelligence system that aggregates, scores, and curates AI security content from multiple sources.

Overview

Private backend aggregates AI security content → scores with multi-dimensional scoring → summarizes with LLM → generates private artifacts

Public GitHub Pages frontend publishes curated daily briefs, trends, and learning resources

Built to support career transition into AI Security Research roles and generate presentation/paper ideas.

Features

🔍 Content Aggregation

• RSS feed fetching with caching and rate limiting
• arXiv API integration for academic papers
• Configurable source credibility tiers
• Deduplication and clustering

📊 Multi-Dimensional Scoring

• Relevance: Keyword matching to AI security priorities
• Credibility: Source authority assessment
• Impact: Severity and affected users
• Freshness: Recency scoring with time decay
• Practicality: Actionable insights detection

Final weighted score (0-100) for intelligent ranking.

🤖 LLM Summarization

• Local Ollama integration (zero API cost)
• Fallback to metadata-based summaries
• Generates: summary, why_it_matters, practical_mitigation
• Grounded prompts prevent hallucination

📈 Automated Workflows

• Daily (7 AM America/Edmonton): Fetch → Process → Score → Summarize → Publish
• Weekly (Sunday 9 AM): Generate PDF briefs, LinkedIn drafts, presentation points, trends

🌐 GitHub Pages Site

• Daily brief with top 10-20 items
• Learning resources (categorized: Start/Intermediate/Deep Dive)
• 30-day trends dashboard
• Professional, mobile-responsive design

Quick Start

# Clone and setup
git clone <repository-url>
cd neo-notebook.github.io
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Configure sources
# Edit config/sources.yaml to add your RSS feeds

# Run daily pipeline
python src/main.py --mode daily

# View outputs
open docs/public/index.html

See docs/setup.md for complete installation guide.

Project Structure

├── .github/workflows/     # GitHub Actions (daily/weekly)
├── config/               # Configuration (sources, scoring, tags)
├── docs/                 # Documentation
│   └── public/          # GitHub Pages site
├── private/             # Private outputs (gitignored)
│   ├── feeds/          # Daily JSON archives
│   └── artifacts/      # Weekly PDFs, LinkedIn drafts
├── src/                 # Python pipeline
│   ├── fetchers/       # RSS/arXiv fetching
│   ├── processors/     # Parse, normalize, dedupe, cluster
│   ├── scoring/        # 5-dimensional scoring
│   ├── summarization/  # LLM + fallback
│   └── outputs/        # JSON, HTML, PDF generators
└── tests/              # 87 tests, 100% passing

Technology Stack

• Backend: Python 3.11+
• LLM: Ollama (local, zero-cost)
• Frontend: Pure HTML/CSS/JS (no build step)
• Automation: GitHub Actions
• Testing: pytest (87 tests)

Configuration

Content Focus

High priority: Agentic security, prompt injection, HITL, observability, shadow AI

Filters: Excludes generic ransomware/breach recaps without technical insights

Scoring Weights

relevance: 0.35      # Topic match
credibility: 0.25    # Source authority
impact: 0.15         # Severity
freshness: 0.15      # Recency
practicality: 0.10   # Actionable insights

Fully configurable in config/scoring_weights.yaml

Documentation

• Setup Guide - Installation and configuration
• Sources - Source list and credibility tiers
• Taxonomy - Tags and scoring rationale

Testing

# Run all tests
PYTHONPATH=. pytest tests/ -v

# Run specific test suite
PYTHONPATH=. pytest tests/test_processors.py -v

Test Coverage: 87 tests covering:

• Processors (35 tests)
• Scoring system (15 tests)
• Summarization (15 tests)
• Output generators (8 tests)
• Utilities (9 tests)
• Integration (5 tests)

License

MIT License

Contributing

Built with Claude Code. Contributions welcome via pull requests.

---

🤖 Generated with Claude Code