LegalTech / AI assistant / Multi-tenant SaaS
Jurisync
Secure legal AI for law firms
Legal research takes hours out of a lawyer’s week, and general-purpose AI assistants answer with confidence but without sources. Jurisync gives law firms a research assistant built for the Canadian legal context: it searches recognized legal sources, cites what it finds in legal citation format, and keeps each firm’s conversations in its own isolated tenant.
- 9 Canadian and Québec legal domains the web search is restricted to
- 18 row-level security policies isolating each firm across 7 tables
- 4 successive searches per question in deep-research mode
- Status
- In production
- Model
- B2B
- Market
- Québec and Canada
The problem
Lawyers and paralegals need answers grounded in case law and legislation, not plausible paragraphs. A general-purpose chatbot does not know which sources are authoritative in Québec and Canada, invents references when it runs out of them, and gives a firm no separation between its users and the next firm’s. Confidential case files raise a second, distinct problem: they should never be pasted into a public assistant at all.
The solution
A multi-tenant SaaS with a public research mode designed for Canadian law: the question goes to a large cloud language model that runs a research loop with tools, a web search restricted to nine Canadian and Québec legal domains, and an answer in a fixed structure — legislative framework, case law, analysis, limits — with every source cited. Each firm becomes a tenant at sign-up; conversations are saved, titled and kept private to their author. Every screen states that the assistant provides general legal information, not legal advice.
What the platform does
- Research loop with tools For each question the model runs an orchestrated loop of up to five tool calls — a web search plus reference tools that direct it to CanLII, the Civil Code of Québec, LégisQuébec and federal legislation — and folds each result back into its answer, streamed to the browser as it is written.
- Web search limited to legal sources The search engine only returns results from nine Canadian and Québec legal domains, with a primary model and a fallback, so the material the assistant reasons from is the material a lawyer would open.
- Deep-research mode For harder questions, four successive searches are run and then synthesized into a single answer, with the citations of every pass collected together.
- Structured, cited answers Answers follow the same structure every time — legislative framework, case law, analysis, limits — and the system prompt forbids invented references. Citations are collected, displayed and copied in Canadian legal citation format.
- Conversation history Conversations are saved with an automatic title, reopened from a list and deleted with confirmation; each one belongs to its author only.
- One tenant per firm A firm’s tenant, profile and default role are created when its first user signs up. Row-level security on all seven tables — 18 policies — separates firms at the database layer, and a role model (administrator, lawyer, paralegal, compliance officer) is defined in the data model.
Architecture
A single-page web application; managed Postgres with row-level security and email-and-password authentication; serverless functions that orchestrate the model calls, run the tool loop and stream answers to the browser; two cloud AI providers — a large language model for reasoning and a citation-oriented search engine — called from the server side only, behind a provider abstraction. Next on the roadmap: a private analysis environment on a local inference node inside the firm, for case files that must not leave it.
Who uses it
Lawyers and paralegals in law firms, each firm working in its own tenant.
Markets
Québec and Canada: system prompts, citation format and legal sources are specific to Canadian and Québec law; the interface is built French-first with English localization.
Where AI fits — and where it does not
Public research runs on cloud AI: the question and the conversation history are processed server-side by a large language model provider and a web-search provider, and the answer comes back streamed, with its sources. That is why the interface tells users not to submit confidential documents in this mode, and why a private analysis environment on a local inference node is the next step on the roadmap. The assistant provides general legal information with citations; it does not replace a lawyer’s judgment, and every reference is meant to be read before it is relied on.
My role
Founder and architect: product vision, platform architecture and technical direction — the research orchestration, the multi-tenant model and the roadmap toward a private analysis environment; my team at SOFT M POWER CANADA built it and operates it under my direction.