Law Firm System
AI drafting under lawyer review, from intake to a court-ready file.
The problem it solved
Law firms lose hours turning scattered client documents into a filing: reading scans, pulling out dates and amounts, finding the legal basis and drafting in the court's form. Generic AI tools draft fast but invent facts and cannot be trusted with a client. This system drafts from the firm's own approved models, refuses any date or amount that is not in the documents, and puts a lawyer's approval between the machine and the client.
Similar problems it can solve
Litigation and immigration practices, in-house legal teams and legal-aid services: petitions, defence briefs, complaints, formal notices, contracts and written consultations; training junior lawyers on delivered documents; any regulated document workflow that needs AI speed with human sign-off and an audit trail.
An end-to-end platform for law firms. Clients open a case and upload documents; a per-service AI pipeline extracts the facts, finds the legal basis in a curated corpus and live research, drafts from lawyer-approved models, a critic agent reviews the draft, and a lawyer approves it on a shared review desk before the client sees it. Around it: case and deadline management, a client portal, billing and subscriptions, and a full audit trail. Deployed for Canadian Federal Court immigration filings (Mandamus, memoranda of argument) and for a civil-law court system.
- Per-service AI pipeline: facts, legal basis, drafting, critic review, lawyer approval
- Anti-hallucination guards: every date and amount traced to the inputs; client-statement and scope checks
- Versioned, evaluated AI releases; staff MFA, audit log and encrypted provider keys
The system
An end-to-end platform for law firms: the client opens a case and uploads documents, an AI pipeline built for that service drafts the filing, and a lawyer approves it before the client sees it. It has run for Canadian Federal Court immigration work (Mandamus demand letters and memoranda of argument for judicial review) and for a civil-law court system.
Case management
- Admin-configurable intake forms per practice area, with conditional questions
- A case file with documents, versions, notes, phases and a status workflow
- A shared lawyer review desk: cases arrive with the reason they were escalated, a lawyer claims a case and keeps it through delivery
- Configurable rules for when a lawyer must review before delivery, and an optional review before drafting starts
- Service-level deadlines with near-due and overdue alerts, staff work queues and an audited archive
- Paid revision rounds after delivery, and corporate clients with a dedicated lawyer per organisation
Documents and AI drafting
- Text extracted from PDF and Word; scans and photos read by a vision model once and cached
- Input checks before any work: volume limits, duplicate removal, handwriting and wrong-document detection, hidden instructions inside an image
- A separate multi-step pipeline per service and tier: fact extraction, legal-basis analysis, drafting from lawyer-approved model documents, critic-agent review, guards, lawyer review
- Petitions, defence briefs, complaints, formal notices, contracts and consultations, delivered as PDF, copyable text and Word
- A lawyer desk with version history and rollback, AI revision on instruction, and correction of machine-read text
- A training edition: every delivered document becomes a section-by-section lesson for junior lawyers, with self-test questions
Research and knowledge
- A curated legal corpus (statutes, precedent rulings, a terminology glossary, model pleadings) fed only to the step that needs it
- Live legal research where every finding carries its source and date, with a guard that drops any query containing a name or case identifier
- Legal updates enter the library only after human approval, and outrank the model's own memory
- Lawyer corrections are scrubbed of personal data and kept for evaluation
Trust, quality and security
- Fact provenance: every date and amount in a draft must appear in the inputs; statements attributed to the client are checked against what the client said
- Prompt-injection defence: document text is data, never instructions, and an output guard stops system instructions leaking into a deliverable
- Versioned AI releases that are snapshotted, evaluated and preview-tested before go-live; model and reasoning depth chosen per step
- Fails closed on a truncated answer; retries only on transient provider failures
- Passwordless client sign-in, mandatory two-factor sign-in for staff, role-based panels, a before/after audit log and encrypted provider keys
Client side
Client portal with a service catalogue, case tracking, notifications by email and SMS, support tickets, invoices, subscriptions with monthly credits, and recorded acceptance of terms. A no-sign-up triage guide recommends the right service.
Technology
.NET 10 · ASP.NET Core · EF Core · SQL Server · multiple LLM providers with vision OCR · retrieval over a legal corpus · QuestPDF · OpenXML