I built Verígrafo (https://verigrafo.com), a legal intelligence platform that makes Spain’s official government bulletins searchable via natural language queries, with verified citations down to the exact document, page, and paragraph.
The problem: Spain publishes 3,500+ pages daily across dozens of official bulletins (BOE, BORME, regional bulletins). Lawyers, civil servants, and businesses depend on this information but the existing search tools are essentially keyword search over PDFs from the 2000s.
The stack:
- Temporal Knowledge Graph: 38.7M nodes, 58M+ relationships built from 300K+ documents (2020–present)
- Graph database: Memgraph for entity/relationship queries
- Vector search: ChromaDB for semantic retrieval
- Zero-hallucination guarantee: every answer cites the exact source location, linked to the official PDF with the passage highlighted
The TKG captures entities (people, organizations, laws, dates), their relationships, and how those relationships change over time. This lets you ask questions like “Who is the current director of [agency]?” or “What regulations modified Law X since 2022?” — queries that pure vector search can’t answer.
Also includes daily summaries organized by bulletin and automated email alerts for new publications matching your criteria.
This week, the city of Las Palmas was nominated for a national award (@aslan) for building a similar RAG + graph system for one municipal department. Verígrafo covers the entire country.
I’m a former NASA JPL / Amazon Robotics / Disney Imagineering engineer who moved back to the Canary Islands and built this solo.
Hey HN,
I built Verígrafo (https://verigrafo.com), a legal intelligence platform that makes Spain’s official government bulletins searchable via natural language queries, with verified citations down to the exact document, page, and paragraph.
The problem: Spain publishes 3,500+ pages daily across dozens of official bulletins (BOE, BORME, regional bulletins). Lawyers, civil servants, and businesses depend on this information but the existing search tools are essentially keyword search over PDFs from the 2000s.
The stack: - Temporal Knowledge Graph: 38.7M nodes, 58M+ relationships built from 300K+ documents (2020–present) - Graph database: Memgraph for entity/relationship queries - Vector search: ChromaDB for semantic retrieval - Zero-hallucination guarantee: every answer cites the exact source location, linked to the official PDF with the passage highlighted
The TKG captures entities (people, organizations, laws, dates), their relationships, and how those relationships change over time. This lets you ask questions like “Who is the current director of [agency]?” or “What regulations modified Law X since 2022?” — queries that pure vector search can’t answer.
Also includes daily summaries organized by bulletin and automated email alerts for new publications matching your criteria.
This week, the city of Las Palmas was nominated for a national award (@aslan) for building a similar RAG + graph system for one municipal department. Verígrafo covers the entire country.
I’m a former NASA JPL / Amazon Robotics / Disney Imagineering engineer who moved back to the Canary Islands and built this solo.
Demo video (3 min): https://www.youtube.com/watch?v=bTYNiAbaH6I
Free tier available (5 queries/day). Would love feedback from the HN community, especially on the graph architecture decisions.