Stop paying lawyers to stay compliant. Vigil monitors EU regulations in real time and delivers personalized, actionable compliance alerts – before the deadline hits.

Built at GenAI Zürich Hackathon 2026 · Powered by Apify · OpenRouter · Claude

🌐 Live demo: https://vigil-demo.streamlit.app/

Vigil is an AI-powered regulatory intelligence tool that helps European SMEs track EU regulatory changes relevant to their business – automatically, in plain language, personalized to their industry and country.

Given a company profile (name, industry, country, regulations of interest), Vigil:

1. Scrapes EUR-Lex, EDPB, GDPR, and 17 national regulatory sources in parallel using Apify's website-content-crawler
2. Chunks documents and extracts structured facts via LLM (Claude 3 Haiku via OpenRouter)
3. Embeds facts using text-embedding-3-small for semantic search
4. Filters and retrieves relevant facts using a two-step hybrid approach: keyword scoring + cosine similarity vector store
5. Generates a personalized monthly digest, deadline alerts, strategic insights, and plain-language summaries
6. Exports the full report as PDF or Markdown

raw_documents (Apify scraping)
    ↓
chunk_documents()         # paragraph-level chunking with overlap
    ↓
extract_facts()           # LLM extracts structured JSON facts per chunk batch
    ↓
embed_facts()             # text-embedding-3-small via OpenRouter
    ↓
filter_relevant()         # hybrid: keyword scoring + severity boost
    ↓
retrieve()                # cosine similarity vector store (in-memory)
    ↓
generate_digest()         # LLM generates monthly digest + plain language summary
generate_alerts()         # keyword detection + LLM-based alert extraction
    ↓
format_output()           # structured JSON → Apify dataset + Markdown/PDF report

• Fact-based embeddings instead of chunk-based: LLM extracts discrete regulatory facts (claim, regulation, article, deadline, action_required, severity) before embedding. This improves retrieval precision significantly.
• Two-step retrieval: keyword pre-filtering with industry/country/regulation-specific keywords, then cosine similarity on the filtered subset.
• 30-day recency window: prompts instruct LLM to only report changes from the last 30 days, filtering out historical regulatory background.
• Parallel scraping: all three scrapers (EUR-Lex, GDPR, national) run via asyncio.gather for 3x speed improvement.

vigil_genAI-hackathon-2026/
├── .actor/
│   ├── actor.json                  # Apify Actor metadata
│   └── input_schema.json           # Actor input schema
├── src/
│   ├── scrapers/
│   │   ├── eurlex_scraper.py       # EUR-Lex + EU Commission news
│   │   ├── gdpr_scraper.py         # EDPB + national DPA sources
│   │   └── national_scraper.py     # 17 EU country regulators
│   ├── processing/
│   │   ├── chunker.py              # Paragraph-level chunking with overlap
│   │   ├── fact_extractor.py       # LLM fact extraction to structured JSON
│   │   ├── embedder.py             # OpenRouter embeddings + fallback
│   │   └── relevance_filter.py     # Hybrid keyword + cosine scoring
│   ├── rag/
│   │   ├── vector_store.py         # In-memory cosine similarity store
│   │   ├── retriever.py            # Two-step retrieval
│   │   └── prompt_templates.py     # All LLM prompts with recency constraints
│   ├── digest/
│   │   ├── digest_generator.py     # Monthly digest + plain language summary
│   │   ├── alert_engine.py         # Keyword + LLM alert detection
│   │   └── formatter.py            # JSON + Markdown + PDF output
│   └── main.py                     # Apify Actor entrypoint
├── app.py                          # Streamlit frontend
├── tests/
│   └── test_vigil.py               # 22 integration tests (real API calls)
├── demo/                           # Screenshots and example reports
├── .streamlit/                     # For cloud Streamlit deployment
├── Dockerfile
├── requirements.txt                # Production dependencies
├── requirements-dev.txt            # Dev + test dependencies
├── vigil_onepager.html
├── README.md
├── .gitignore
└── .env.example

• Python 3.12
• Apify account with API token
• Apify CLI (for Actor deployment)

git clone https://github.com/allesgrau/vigil_genAI-hackathon-2026.git
cd vigil_genAI-hackathon-2026

conda create -n vigil python=3.12
conda activate vigil

# Production only (for Apify Actor)
pip install -r requirements.txt

# Development + Streamlit + tests
pip install -r requirements-dev.txt

cp .env.example .env

Edit .env:

APIFY_TOKEN=your_apify_token_here
OPENROUTER_ACTOR_URL=https://openrouter.apify.actor/api/v1

Get your Apify token at: https://console.apify.com/account/integrations

streamlit run app.py

Open http://localhost:8501

python src/main.py

Tests make real API calls (Apify + OpenRouter) with minimal data – same philosophy as test_mode. Expected cost: ~$0.05 per full run.

pytest tests/test_vigil.py -v

Test coverage: chunker, fact extractor, embedder, relevance filter, digest generator (including no-past-deadlines assertion), formatter.

apify push

Then publish the Actor from https://console.apify.com/actors

Actor input schema:

{
  "company_name": "TechStartup GmbH",
  "industry": "fintech",
  "country": "DE",
  "size": "startup",
  "areas_of_concern": ["GDPR", "AI Act", "DORA"],
  "test_mode": true
}

Supported industries: fintech, healthcare, ecommerce, saas, manufacturing

Supported countries: DE, PL, FR, CH, NL, IT, ES, AT, BE, SE, IE, LU, DK, FI, PT, CZ, HU, RO

Supported regulations: GDPR, AI Act, PSD2, AML, NIS2, DORA

test_mode: true – scrapes 2 URLs, max 3 pages per source (~$0.10/run)

test_mode: false – full scraping, 15 pages per source, supports link depth = 1 (~$10/run)

EU-level:

• EUR-Lex Official Journal (eur-lex.europa.eu/oj/direct-access.html)
• EU Commission press corner
• EDPB news (edpb.europa.eu/news/news_en)
• GDPR.eu news
• Regulation-specific pages (DORA, NIS2, AI Act, PSD2, AML)

National regulators (17 countries):

Vigil outputs a structured JSON object saved to the Apify dataset:

{
  "generated_at": "2026-03-16T11:01:00Z",
  "company_name": "Dunkelblau",
  "industry": "fintech",
  "country": "DE",
  "sources_analyzed": 10,
  "alerts_count": 2,
  "critical_alerts": [...],
  "high_alerts": [...],
  "medium_alerts": [...],
  "digest_markdown": "...",
  "plain_summaries": [...],
  "full_report_markdown": "...",
  "status": "success",
  "vigil_version": "0.1.0"
}

Vigil is a regulatory intelligence tool, not legal advice. Always consult a qualified legal professional for compliance decisions.

Built at GenAI Zürich Hackathon 2026 🇨🇭