Comprehensive Project Synopsis
Inspiration
ClaimReady began with a simple problem: self-represented claimants are increasingly able to use generative AI when preparing legal claims, but ordinary AI tools are not designed for the discipline required in court proceedings.
A general-purpose chatbot may produce information that sounds convincing but is unsupported, omit facts that matter to the claim, or accept the claimant's assumptions without testing them. In a legal setting, those weaknesses are serious because the user may rely on the output when completing a court form.
Our starting question was therefore not simply, “How can AI make claim preparation faster?” It was: “How can AI assist a self-represented claimant without encouraging them to trust information that cannot be verified?”
That led to the principle that governs ClaimReady: trust before speed.
We designed the project for Singapore's Small Claims Tribunals and anchored it to the Singapore Courts' Guide on the Use of Generative AI Tools by Court Users. Rather than treating responsible AI as an additional feature, we used that concern to shape the product itself.
What it does
ClaimReady is an AI-assisted companion for people preparing their own Small Claims Tribunal claims.
It is designed as a self-contained overlay on the SCT pre-filing page, rather than as a separate, open-ended chatbot. The aim is to assist the claimant at the point where they are already answering the Tribunal's questions.
Its present working version focuses on two parts of the pre-filing process:
Route triage. ClaimReady guides the user through the relevant preliminary questions so that the information required by the SCT process is considered in a structured order.
Auto-filled pre-filing assistance. The claimant may upload a PDF, Markdown file or plain-text document, or describe what happened directly in a text field. ClaimReady searches that material for information relevant to each SCT question and proposes an answer where the evidence supports one.
The important feature is that ClaimReady does not simply return an answer. Each suggestion is designed to show:
the proposed answer;
the document and page supporting it;
the relevant quotation from the source; and
a confidence score based on fit, evidence, retrieval, consistency and source reliability.
For example, the deck demonstrates a suggested payment amount of $500, traced to a payment receipt stating “Total paid: $500.00”, with the individual components of the confidence score exposed to the user.
Where the evidence is insufficient, ClaimReady is intended to withhold the suggestion rather than guess.
The system also constrains the answers it can propose. Its extraction process may be probabilistic, but the possible answers are bounded by the actual questions and answer choices used on the SCT website. In the language of the deck: the process is probabilistic; the set of possible answers is not.
ClaimReady also contains a two-stage tone and bias check. A fast local check identifies obviously hostile or generalising language. Where necessary, a deeper language-model review considers the wording in context. Instead of simply endorsing the claimant's language, ClaimReady can point out problematic phrasing and suggest a more factual formulation. The user remains free to reject the suggestion.
The wider product is conceived as one continuous journey from initial claim preparation to post-filing guidance. At present, however, the deck is careful to distinguish what exists from what is planned:
Live: route triage and the auto-filled pre-filing form.
Planned: validated submission, a generated reference PDF and an after-filing checklist.
How we built it
ClaimReady separates the user interface from the processing layer.
The frontend is a lightweight, self-contained JavaScript overlay. It is designed to sit over the existing pre-filing experience without depending on the underlying host site.
The backend uses FastAPI and supports the main processing functions, including tone detection, OCR for scanned documents and citation verification.
Log in or sign up for Devpost to join the conversation.