CivicLease NSW

Source-backed rental issue triage for NSW students and renters.

Comment

Inspiration

Rental problems often begin as one stressful message: an agent wants to keep part of a bond, a repair keeps getting delayed, rent goes up, someone wants to inspect the property, or a tenant is told to leave. For a student renter, the first hard step is usually understanding what kind of issue this is and which official NSW source to check.

What it does

CivicLease NSW turns a plain-language rental problem into structured triage:

  • primary issue category
  • urgency and confidence
  • dates, amounts, and evidence cues
  • official NSW Government, NSW Fair Trading, and NCAT source cards
  • recommended next steps and evidence checklist
  • uncertainty and safety disclaimer

It also abstains when the issue appears unsupported, outside NSW residential renting, or written from a landlord-side workflow.

How we built it

The prototype is a static browser app built with HTML, CSS, and JavaScript. The analysis pipeline combines an explainable Naive Bayes classifier, category cue signals, deterministic fact extraction, urgency detection, role/scope guards, and source routing from a local sources registry.

No LLM API is used in the submitted MVP. The user's text stays in the browser.

Evaluation

CivicLease correctly classified 119 of 120 cases (99.2%) on our author-labelled MVP evaluation set, compared with 61/120 (50.8%) for the keyword baseline.

This is a small hackathon-scale evaluation, not a production or research benchmark. Known limitations remain: labels need independent review, edge cases may still abstain or misclassify, and users should verify official sources or seek qualified legal help.

Challenges

The main challenge was keeping the product useful without pretending to be a lawyer. I had to make the output source-backed, explainable, privacy-friendly, and honest about uncertainty. Another challenge was multi-issue ranking, because a renter may mention arrears, repairs, entry, and bond problems in the same paragraph.

Accomplishments

  • Built a working local browser prototype.
  • Added official NSW/NCAT source routing.
  • Added abstention for unsupported and landlord-side inputs.
  • Built a 120-case MVP evaluation set and a separate 20-case post-tuning challenge set.
  • Prepared demo screenshots, script, backup scenario, and recording checklist.

What I learned

For legal and civic tools, the valuable AI task is often not final advice. It is safer triage: identify the issue, show the source, expose uncertainty, and help the user prepare the next conversation.

What's next

  • Add lawyer or tenant-advocate review of labels and outputs.
  • Add multi-label summaries for cases with several active issues.
  • Add source freshness checks.
  • Add exportable evidence checklists.

Disclosure

OpenAI Codex and ChatGPT assisted with planning, code generation, documentation, testing, and review. The submitted MVP itself does not call an LLM API.

Disclaimer

CivicLease NSW provides legal information and issue triage for NSW residential renting. It is not legal advice and does not replace qualified legal help.

Built With

Share this project:

Updates

Submission history