Québec legal research, grounded in a knowledge graph.

The problem

The Québec legal market is poorly served. Common-law research tools are not built to model Québec’s civil-law corpus at the article level. US platforms are billed per user, in US dollars, on a monthly subscription, and reason in common law. Public portals give the text of the statutes, with no synthesis and no reasoning.

None of these tools is built on the full body of Québec and federal legislation and regulation, indexed article by article, with the cross-references between provisions modelled as relations.

How it works

Jurisgraph runs on a Neo4j knowledge graph: roughly 37,320 provisions drawn from 248+ statutes and 6,459 regulations, Québec and federal, indexed article by article, with the cross-references between provisions modelled as relations. A LangGraph orchestration chains graph retrieval and vector retrieval, a reranker fine-tuned on Québec law, deterministic filtering, then structured legal reasoning following the IRAC method (Issue, Rule, Application, Conclusion). Every answer carries a reasoning certificate: the sources used, the validation chain, and a confidence score from 0 to 100.

What sets jurisgraph apart

Built for Québec

Québec is a civil law jurisdiction, in French, with its own body of law. Jurisgraph is not a common-law tool retrofitted after the fact.

• Native civil law Modelled on the Civil Code of Québec and the RLRQ corpus, not transposed from common law.
• French first Reasoning and answers are produced in French legal register; English is available.
• Data residency Designed for Québec data residency and Law 25 from the ground up. The target architecture is hosting in Canada.
• Official sources Sourced from LégisQuébec and the Statutes of Canada — legislation and regulation, kept current.

Who it’s for

Jurisgraph is designed for people who need to ground a position in the exact text of the law. The private beta is aimed at these roles first.

• Solo practitioners and small firms

  Cover more ground without a research team: one sourced answer, verifiable provision by provision.

• Litigation

  Quickly map the applicable provisions and their cross-references before drafting a pleading.

• In-house counsel

  Answer internal questions with a trace of the sources, not an unsupported note.

• Compliance and privacy

  Tie each requirement to its precise legal basis, with the validation chain attached.

• Academic work

  Explore the regulatory corpus as a graph: follow the cross-references between provisions step by step.

Proof

The product exists and it is measured. These are the numbers a program like AWS Activate or Neo4j Startup looks at to tell a real product from a promise.

The fine-tuned reranker (bge-reranker-v2-m3 + LoRA, 7,258 pairs) reaches NDCG@5 ≈ 0.99 on an internal held-out test set. This result is not published and not independently verified. Jurisgraph is in private beta: no clients or deployments to date.

Roadmap

• Streaming answers
  Reasoning appears as it is produced rather than in a single block.
• Word and Outlook integration
  Use jurisgraph where lawyers actually draft.
• Certification path
  Formalise the security posture, with a SOC 2-type certification as the goal.
• Corpus expansion
  Beyond legislation and regulation, subject to licensing agreements.

Team

Built on

Neo4j first: the knowledge graph is the core of the product.

• Neo4j
• LangGraph — pipeline orchestration
• Python · FastAPI
• Fine-tuned reranker — bge-reranker-v2-m3 + LoRA
• Claude Sonnet — synthesis and reasoning
• Open models via Groq — fast deterministic steps
• OpenAI embeddings
• AWS — target hosting

Frequently asked questions

Request a demo

Tell us briefly about your context. We answer requests one by one, with pilot access where the fit is right.