Contract Management System

AI-powered contract review management system that checks your clauses against real Australian legislation

Comment

ClauseCheck

Inspiration

Every year, hundreds of thousands of residential and retail leases are signed across NSW. Most tenants and small business owners sign without fully understanding whether the terms are legally compliant — and a single consultation with a property lawyer costs $300–$800. On the other side, property lawyers spend 45–90 minutes manually reviewing each tenancy contract clause-by-clause against the Residential Tenancies Act 2010, for fixed-fee matters that might only bill $250–$600. It's the most time-intensive, lowest-margin work in a property law practice.

We experienced this firsthand. As UNSW students renting in Sydney, we reviewed friends' tenancy agreements and found multiple void clauses being enforced — landlords holding bonds without lodging them with Fair Trading, illegal electricity cap arrangements, and notice periods that didn't match the Act. We realised that if two university students could catch these issues, a purpose-built AI tool could do it faster, more thoroughly, and at scale.

The second problem we noticed: even after a lawyer reviews a contract and sends a response to the landlord, tracking what was agreed, disputed, or ignored across multiple clauses becomes a mess of email threads and sticky notes. There's no structured system for managing the negotiation lifecycle of a lease.

That's why we built ClauseCheck — not just a contract reviewer, but a full workflow tool that goes from upload to resolution.

What it does

ClauseCheck is an AI-powered lease contract review and correspondence management system for NSW residential and retail leases. The workflow has four stages:

  1. Upload — Drop in any lease PDF (including scanned documents). The system extracts and segments every discrete clause.
  2. AI Review — Each clause is benchmarked against the relevant NSW legislation and classified as Invalid (void under the Act), Arguable (context-dependent), or Valid, with exact section citations and plain-English reasoning.
  3. Respond — The lawyer reviews each finding in an editorial interface — accepting, dismissing, or editing each verdict and adding their own notes. Accepted findings auto-generate a professional correspondence letter to the landlord citing the specific Act sections, ready to review and send.
  4. Track — After sending, the lawyer can paste the landlord's reply and AI analyses whether each concern was agreed to, disputed, or ignored. The dashboard tracks the status of every flagged clause across all contracts.

Legislation covered

  • Residential Tenancies Act 2010 (NSW)
  • Residential Tenancies Amendment Act 2024 (NSW)
  • Residential Tenancies Regulation 2019 (NSW)
  • Retail Leases Act 1994 (NSW)
  • Retail Leases Regulation 2022 (NSW)
  • Competition and Consumer Act 2010 (Cth) — Australian Consumer Law, unfair contract terms provisions
  • Anti-Discrimination Act 1977 (NSW)
  • Strata Schemes Management Act 2015 (NSW)
  • Property and Stock Agents Act 2002 (NSW)

How we built it

Frontend: React with Tailwind CSS. The interface includes a landing page, contract upload with PDF parsing, an inline clause review editor with accept/dismiss/edit workflow per finding, email generation and preview, and a contract dashboard with status tracking across all reviewed contracts.

Backend: Python with FastAPI. Handles PDF ingestion, clause segmentation, and orchestrates the AI analysis pipeline.

AI Pipeline: We use Google Gemini 2.5 Flash for the core analysis. The pipeline works in stages:

  • A local keyword-matching system identifies which legislation files are relevant to the uploaded contract (residential vs. retail vs. mixed)
  • The relevant legislation is loaded from our structured knowledge base — the full text of each Act, encoded and indexed
  • Gemini analyses each clause against the applicable sections, producing structured JSON output with risk level, verdict, section citations, plain-English reasoning, and recommendations
  • Email generation uses the accepted findings to compose a professional letter with statutory references

Knowledge Base: We built a structured repository of NSW legislation specifically relevant to tenancy and retail lease review. Each Act is stored as a searchable document that gets injected into the AI context alongside the contract being reviewed.

Challenges we faced

Token limits and rate limiting. Our initial approach sent the full text of all relevant legislation (up to 5MB) in a single API call. This immediately hit Gemini's 1M tokens-per-minute rate limit. We solved this by implementing intelligent legislation selection (only sending the 2-3 most relevant Acts per contract type) and planning a legislation trimming pipeline to strip administrative boilerplate from the source files.

Analysis speed. The first version took 4-5 minutes per contract. We reduced this by eliminating an unnecessary API call for legislation classification (replaced with local keyword matching), removing an artificial sleep delay, and adding PDF hash caching for instant results on repeat uploads.

Balancing legal precision with AI limitations. The tool needs to be accurate enough for lawyers to trust, but we can't have it giving legal advice. We solved this by positioning every output as a "finding" with citations that a solicitor reviews and signs off on — the AI flags, the human decides. This also addresses the regulatory boundary between legal information and legal advice.

Handling informal and non-standard leases. Many real-world tenancy agreements in NSW aren't the standard Fair Trading form — they're informal letters, emails, or cobbled-together documents. Our clause segmentation needed to handle messy, unstructured input, not just clean legal templates.

What we learned

  • NSW tenancy law is layered but tractable — for residential and retail leases specifically, there's a clear primary Act plus a shared federal overlay (ACL unfair contract terms), which makes automated compliance checking feasible
  • The real product insight: contract review tools that stop at "here are the issues" miss the workflow. Lawyers need to act on findings, communicate them, and track responses. The full loop is what makes this a daily-use tool, not a novelty
  • Customer discovery matters more than code. Conversations with 27 practicing property lawyers and 14 law students shaped every product decision, from the B2B positioning to the "lawyer-in-the-loop" design
  • Shipping fast means making hard scoping decisions — we focused on NSW residential and retail leases rather than trying to cover every contract type in Australia

What's next

  • First law firm pilot: 10 seats, 4 weeks free, measuring time saved per review vs. baseline
  • UNSW Legal Clinic partnership as an institutional pilot
  • Legislation trimming to reduce analysis time further
  • Expand to VIC Residential Tenancies Act and QLD legislation
  • Integration with legal practice management software (LEAP and Smokeball)

Built With

Share this project:

Updates