SB
8M+ regulatory nodes

Eryndal

Enterprise RAG across CVEs, GDPR, DORA, AI Act, and NIS2. Semantic and structural vectorisation with hierarchical chunking that respects legal document structure.

A regulatory question becomes a vector — a dense numerical representation of meaning. That vector is compared against millions of stored document chunks in embedding space. The nearest chunks surface not by keyword match, but by semantic proximity. This is fundamentally different from traditional search.

Select a query to see retrieval in action
Embedding space
Select a query to begin
Top-k
Speed
Watch Eryndal in action →

Naive text splitting every 512 tokens destroys the structure that makes regulations interpretable. An article split across two chunks loses its context. Eryndal chunks with legal structure awareness — each piece knows its parent article, chapter, and regulation. The difference between usable retrieval and broken context.

Toggle between naive and hierarchical chunking
Document structure
Resulting chunks
Art. 25, Para 1✓
DORA > Ch. V > Art. 25 > Para 1
Art. 25, Para 2✓
DORA > Ch. V > Art. 25 > Para 2
Art. 25, Para 3✓
DORA > Ch. V > Art. 25 > Para 3
Art. 26, Para 1✓
DORA > Ch. V > Art. 26 > Para 1
Art. 26, Para 2✓
DORA > Ch. V > Art. 26 > Para 2
Every chunk knows its parent article, chapter, and regulation
← All projects