Secure, citation-based legal AI system

This is the application that help professionals work with complex information more efficiently. This project focuses on Czech labour law and combines retrieval, structured reasoning, and secure document handling to assist legal research and document analysis.

LinkedIn: marina-mulyukina · Questions or feedback? Feel free to reach out.

Who it's for

This product is designed for legal professionals and teams who need faster, more traceable research across legal sources and private documents.

  • Law firms
  • Employment lawyers
  • Legal departments
  • HR and compliance professionals
  • Legal researchers
  • Students

Problems it solves

The assistant is built to reduce preliminary research time and help legal teams organise complex information into structured, reviewable answers.

  • Reduce time spent on preliminary legal research.
  • Quickly locate relevant statutes and court decisions.
  • Search across firm-specific documents conversationally.
  • Identify missing facts and potential legal risks.
  • Organise legal information into structured, reviewable answers.

Key features

The product surfaces the information lawyers need in business terms, not just technical output.

  • Citation-based answers
  • Private document analysis
  • Conversational search across uploaded files
  • Risk assessment and missing information detection
  • Structured legal summaries
  • Chat history and document-scoped conversations

Security and confidentiality

Confidentiality is a core requirement for legal work, so the system is designed to keep user data private and isolated.

  • User documents remain private and isolated.
  • Uploaded materials are only accessible to their owners.
  • The system does not train on user data.
  • Authentication and ownership checks restrict access.
  • Documents are used solely for retrieval and inference.

Example use cases

Lawyers and legal teams can use the assistant for practical day-to-day research and document review tasks.

  • Research employee termination requirements.
  • Analyse employment contracts and internal policies.
  • Search across uploaded case files.
  • Compare court decisions related to a legal issue.
  • Prepare preliminary legal summaries before client meetings.

Why this is different

This is not a generic ChatGPT wrapper. It is a focused legal knowledge system designed for Czech labour law and professional review workflows.

  • Built specifically for Czech labour law.
  • Uses statutes, case law, and legal concepts as knowledge sources.
  • Produces structured, cited responses.
  • Supports secure private document collections.
  • Designed and tested as a professional knowledge system.

How it works

The workflow is intentionally simple for the user and detailed behind the scenes.

  • Ask a legal question or upload a private document.
  • The system retrieves relevant sources from the legal corpus and your own files.
  • You get a structured answer with citations, risk signals, and missing facts.

How the system works under the hood

The technical stack supports the legal workflow by finding the right source even when the user does not use exact statutory wording.

Question expansion

The user question is rewritten into several search variants to improve recall when wording is incomplete or paraphrased.

Exact legal search

Keyword search over the legal corpus makes it fast to find exact section numbers, statute names, and legal phrases.

Meaning-based retrieval

The assistant understands legal questions even when exact statutory wording is not used, helping retrieve conceptually related legal information.

Linked legal context

The assistant follows relationships between statutes, concepts, and court decisions to surface relevant context that keyword search alone often misses.

Ranked results

Retrieved chunks are ranked so the most relevant, authoritative, and context-rich sources appear first.

Structured answers

The assistant returns a conclusion, risk assessment, missing facts, next steps, and citations in a reviewable format.

Trust indicators

The project is designed to be understandable, testable, and easier to review than a generic AI chat interface.

  • End-to-end system designed and implemented by the author.
  • Automated evaluation and testing pipeline.
  • Citation-driven answer generation.
  • Structured outputs with risk assessment.
  • Focus on reliability and traceability.

Free access and limits

This is a free legal application in its current stage. To keep the service sustainable, stable, and available for everyone, usage limits are enabled in the main chat.

  • Up to 20 answers per chat session.
  • Up to 2 saved chats per user.
  • These limits help control AI infrastructure costs and keep response quality consistent.

Need unlimited or higher-volume usage? A full paid package is available for professional teams that need larger limits, longer sessions, and expanded support.

Technology stack

FrontendNext.js · React · TypeScript · NextAuth
BackendFastAPI · Python · Pydantic · Uvicorn
DataSQLite · SQLite FTS · Memgraph
AI / RetrievalRAG pipeline · AI agents · OpenAI · Memgraph vector index
Testing / EvaluationVitest · Japa · Playwright · Pytest · DeepEval
CI / AutomationGitHub Actions
DeploymentDocker Compose

Get in touch

Interested in exploring legal AI for your firm? Looking for a secure legal research assistant or private document analysis workflow? Feel free to reach out.

LinkedIn: marina-mulyukina