Grading Rubrics

Download the Grading Rubrics here

Problem Statements

Submissions close on Saturday, 13 September 2130H!

Submissions Closed!

About the Sponsor

The Public Defender’s Office (PDO) was established to enhance access to justice to vulnerable persons through the provision of criminal defence aid. Set up as a department under the Ministry of Law, the PDO is led by the Chief Public Defender and staffed by a team of full-time public defenders. Singapore Citizens and Permanent Residents, who are charged with non-capital criminal offences and are unable to afford legal representation, may apply for criminal defence aid with the PDO.

Background

Criminal defence aid application processes face a fundamental accessibility challenge: they must remain available through hardcopy forms to ensure non-tech savvy applicants can access legal support. Yet, this creates significant workflow inefficiencies for government officers processing these applications.

The Current Reality

When applicants fill out criminal defence aid forms, they rarely follow the structured format that digital systems require. Instead of neat, compliant responses in designated fields, officers regularly encounter:

Informal language responses: Applicants write "not sure," "don't remember," or "maybe" in fields that require specific categorical answers
Structural non-compliance: Applicants ignore the intended form structure entirely, writing lengthy narratives in single boxes instead of distributing information across designated fields
Incomplete mandatory fields: Critical information may be missing entirely,struck out, or buried within unrelated text sections
Handwriting interpretation challenges: Officers must decipher varying handwriting quality while simultaneously translating informal responses into system-compliant terminology

The Workflow Friction

This creates a labour-intensive manual interpretation process where officers must:

Decode handwritten text across varying legibility levels
Hunt for mandatory information scattered throughout narrative responses or written in unexpected form sections
Translate informal language into standardised system terminology (e.g., converting "not sure" into "unknown" for dropdown validation)
Make judgment calls about ambiguous or contradictory responses
Handle validation errors when informal responses don't meet system field requirements

Each application requires significant manual processing time, and the interpretation process introduces potential for transcription errors and inconsistent data entry across different officers.

Challenge

Your task is to build an intelligent OCR solution that can handle the messy reality of how people actually complete forms, not just the ideal scenario of neat, compliant responses.

Core functionality

The tool must:

Recognise handwritten text from hardcopy application forms
Convert handwriting into editable and searchable text data
Output data in two formats:
- Downloadable CSV file
- Direct database transfer for front-end case management portal viewing

Real-World Complexity

Your solution must intelligently handle non-ideal applicant behaviour, including:

Responses that ignore form structure and field boundaries
Informal language that doesn't match system validation requirements
Missing or incomplete mandatory field information
Ambiguous responses that require interpretation or flagging for human review

Reading Materials

PDO website
Establishment of the Public Defender’s Office with the introduction of the Public Defenders Bill
Application Process and Eligibility Criteria for Criminal Defence Aid (video)
Application form (online version)
Excerpt of hardcopy application form (refer to image below)

Possible Solutions

An effective solution will reduce the manual interpretation burden on government officers while maintaining the accessibility that hardcopy forms provide to applicants who cannot or prefer not to use digital channels. Consider how your tool might intelligently map informal responses to system-compliant formats, flag uncertain interpretations for human review, and ensure mandatory fields are populated even when applicants provide information in unexpected ways.

Track Prize

Earn the chance to showcase your prototype in a joint presentation to senior leaders at the Ministry of Law.

About the Sponsor

The Legal Aid Bureau(“LAB”) ensures equal access to justice to persons of limited means, by providing them legal aid, advice and assistance in civil matters. This includes representing applicants in many kinds of court proceedings, as set out in the Legal Aid and Advice Act. LAB cases are handled by LAB’s in-house lawyers and assigned solicitors. To qualify for legal aid, an applicant must satisfy both the Means and Merits tests.

Background

eBantu Formula Updater – eBantu is an automated form-filler to help Syariah Court litigants fill in the forms to start their court proceedings. Part of the form involves stating how much you want for nafkah iddah (which is wife maintenance for 3 months, post-divorce), and mutaah (a consolatory gift). e-Bantu can assist to help litigants calculate how much they can get for these items, based on formulae derived from Syariah Court caselaw. The current process of updating eBantu's nafkah iddah and mutaah calculation formulae requires manual review of Syariah Court cases from Lawnet, data extraction, and formula recalibration. This manual approach is resource-intensive, time-consuming, and may lead to delays in reflecting the latest court trends in the eBantu tool. You can access eBantu here.

Challenge

How might we create an automated solution that can systematically extract and analyse relevant case data from Lawnet, generate updated reference tables, and recalibrate the formulae for nafkah iddah and mutaah calculations while maintaining the existing filtering criteria (excluding consent orders, high-income cases, and outliers)?

Reading Materials

PowerPoint Slides here

Possible Solutions

The solution should ideally include all of the following features:

Auto-identification of relevant latest syariah court cases on LawNet (note: optional as there may be technical restrictions with auto-pulling LawNet data, and LawNet is the only source of syariah court case law other than printed law reports)
Auto-extraction of relevant fields (e.g. husband’s income, amount of nafkah iddah and mutaah awarded) from cases
Auto-population of these fields into a spreadsheet
Auto-identification of cases that do not fit the filtering criteria – i.e. excluding consent orders, high-income cases and outliers. Ideally LAB should be able to modify the income threshold considered as “high income” and review which cases the tool has flagged as “outliers”
Auto-updating of nafkah iddah and mutaah formulae based on data from cases that pass the filtering criteria

Track Prize

$50 Grab voucher per participant for the winning team.
Opportunity to present to LAB, enjoy lunch with our Director of Legal Aid and legal tech officers, and receive exclusive LAB goodie bags.

About the Sponsor

AITHENA unites lawyers and engineers to build AI-powered tools, designed by lawyers, for lawyers, that simplify research, streamline workflows to make legal work smarter, faster, and more human.

Background

In every area of legal practice, statutes form the backbone of legal analysis. Whether advising a corporate client on a merger, defending an accused in court, or drafting a privacy policy for a tech startup, a lawyer’s first step and guiding star is often to identify the relevant statutory provisions that apply to a situation.

However, this essential task of identifying relevant statute and applicable sections as well as subsidiary legislation and its regulations can be time consuming, especially in complex situations that engage multiple legal domains, or when dealing with highly regulated matters, like financial derivatives under the Securities and Futures Act.

The process of identifying the statutes applicable to a situation is still a highly manual process for most lawyers, relying on pre-existing domain knowledge to know where to look, and reading of titles of statutes and its provisions. Further, a significant amount of time is often spent cross-referencing within the provisions, schedules and subsidiary legislation of a statute and across related legislation.

Challenge

This hackathon challenges participants to build a tool that can, when given a factual scenario, recommend the most applicable statutory provisions a lawyer must consider. Given the time constraints of this Hackathon, the challenge will be limited to building a tool that can achieve accurate retrieval of relevant statutory provisions in the context of the Personal Data Protection Act 2012, including its schedules and subsidiary legislation.

By automating the retrieval of relevant legislation and subsidiary legislation, such a tool would save lawyers time in the initial canvassing of the legal landscape their case will be operating in, and improve accuracy by reducing the risk of overlooking relevant statutes or cross-references.

Further, this tool would serve as a key building block for more a more advanced AI-driven legal tool that can combine statute retrieval with case law analysis, soft law guidance, and reasoning capabilities. Statute retrieval is not just a technical exercise. It goes to the heart of how lawyers practice law. A tool that can “think like a lawyer” in mapping facts to the right statutes and understand the nuance of cross reference and interaction between statutory provisions and across statutes has the potential to transform legal research and practice.

Reading Materials

Possible Solutions / Technical Guidance

This can be framed as a text classification or a document ranking problem. You could train a model to classify a given scenario into sub areas of data protection (e.g., "Data Breach," "International Data Transfer", “Data Subject Rights Management”, etc) to narrow down the search.
Techniques for Named Entity Recognition (NER) could be used to identify key legal terms, parties, and actions within the scenario description to better match them to statutory provisions.
Retrieval-Augmented Generation (RAG) solution that first performs a semantic search on a vectorised version of the data protection act to find the most relevant sections for the given scenario. Other than similarity based retrieval, there can be variants of RAG also. The Large Language Model (LLM) then uses these specific, retrieved sections as direct source material to generate a precise and context-aware answer.

Track Prize

S$200 and opportunity to work with the team at AITHENA to gain first hand experience at a legal tech start up!

About the Sponsor

Founded in 2019 and recognised as a mission-led company, Jus Mundi is a pioneer in the legal technology industry dedicated to powering global justice through artificial intelligence. Headquartered in Paris, with additional offices in New York, London, and Singapore, Jus Mundi serves over 150,000 users from law firms, multinational corporations, governmental bodies, and academic institutions in more than 80 countries. Through its proprietary AI technology, Jus Mundi provides global legal intelligence, data-driven arbitration professional selection, and business development services.

Background

🌍 Real World Problem
In the evolving landscape of international arbitration, lawyers are expected to synthesize massive documents and records, apply the law, and construct arguments that stand up before tribunals.
Today, some AI tools can answer basic legal questions. But almost none can test the strength of a legal strategy, deconstruct an argument’s logic, find precedent based weaknesses, and suggest superior alternatives. That’s where you come in. 🫵

⚖️ But first: what is arbitration and why should you care?
Arbitration is how the world’s biggest disputes are resolved: globally, and without courts.
When companies clash with governments or when massive infrastructure projects get delayed due to unforeseen crises (like cyberattacks, pandemics, or political turmoil), they don’t go to public courts; they go to arbitration tribunals. These are legal battles, where decisions are enforceable worldwide and billions of dollars, reputations and diplomatic relationships are on the line. ⚓ One weak claim theory can sink all of it.

Challenge

Your mission is to build an LLM-powered solution that acts as a strategic closing statement co-counsel for arbitration lawyers. It’ll be able to help a user on a complex legal task: preparing a persuasive closing statement for an arbitration hearing. This means the user will submit their strategy and some relevant information on their case and AI will review the user’s legal strategy, find factual and legal weaknesses, test it against jurisprudence, and recommend the best closing statement they could draft out of the different findings.

Choose this legal scenario to build around for your presentation

🌿 Scenario – Environmental Counterclaim Risk

You are assisting a legal team representing the Fenoscadia Limited, a large mining company doing its activities in Republic of Kronos. Fenoscadia and Kronos have signed a concession agreement allowing it to exploit a site in Kronos for 80 years.

In 2015, Kronos passed a new Environmental Act (KEA), requiring companies to protect local waters from toxic waste. In 2016, following a government-funded university study suggesting that Fenoscadia’s work may have contaminated the Rhea River (Kronos' main water source), Kronos issued Presidential Decree No. 2424. This decree revoked Fenoscadia’s license, terminated the concession agreement, and banned lindoro exploitation altogether.

A dispute arose.

Fast forward to the middle of the arbitration proceedings, Kronos is seeking to file an environmental counterclaim against Fenoscadia during arbitration proceedings, alleging:

Contamination of the Rhea River due to graspel, a toxic component released during mining;
Health impacts, including increased cardiovascular disease rates;
Costs of cleanup, alternative water supplies, and healthcare, estimated at USD 150 million.

An environmental counterclaim is essentially the host state’s legal response claiming that the investor caused environmental damage, such as pollution or deforestation. The state uses this counterclaim to avoid paying compensation for the investor’s original claims of unfair treatment and losses.

👉 The legal team’s mission is to prepare a closing statement for Fenoscadia, aimed at persuading the tribunal to reject the environmental counterclaim.

Example of a prompt that a user can submit to the assistant:

I’m working on a case representing Fenoscadia Limited, a mining company from Ticadia that was operating in Kronos under an 80-year concession to extract lindoro, a rare earth metal. In 2016, Kronos passed a decree that revoked Fenoscadia’s license and terminated the concession agreement, citing environmental concerns. The government had funded a study that suggested lindoro mining contaminated the Rhea River and caused health issues, although the study didn’t conclusively prove this. Kronos is now filing an environmental counterclaim in the ongoing arbitration, seeking at least USD 150 million for environmental damage, health costs, and water purification.

Can you help me analyze how to challenge Kronos’s environmental counterclaim, especially in terms of jurisdiction, admissibility, and merits?

Example Solution(s)

💡 Your solution MUST be able to do the following

🎁 BONUS (not mandatory, but you can choose 1 or 2 things to implement additionally)

🏆 Why This Challenge Is Special

You’ll have access to 🏆 real legal data. It will be shared the day of the hackathon with thorough guidelines.
Our 🏆 product & legal mentors will bring deep user insight, legal expertise, and real use cases.
Our 🏆 tech mentors will support you with deep technical guidance.
You’ll be having a 🧱 foundational impact for legal techs.
🤖 For Hackers interested in AI Agents, NLP, Solving complex legal problems.

Data Access

Participants will receive data from the Jus Mundi database which can be downloaded here.

See Chris for the password!

Track Prize

500 Euros Cash Prize

About the Sponsor

Ministry of Law (MinLaw) – Legal Technology Transformation Office (LTTO) is a division within the Legal Industry Group that develops strategies and implements initiatives to strengthen Singapore’s status as a global legal services hub through technology. LTTO coordinates and drives legaltech efforts across the legal services sector, promoting innovation and responsible adoption of emerging technologies such as Generative AI.

Background

Generative AI (GenAI) tools are becoming increasingly accessible and sophisticated. Legal professionals are leveraging these tools to augment their work (e.g. conducting legal research, building case chronologies, drafting documents). At the same time, non-legal professionals, including business owners and members of the public, are turning to GenAI for assistance with legal questions, processes, and even drafting their own legal documents.

While these tools offer speed, convenience, and efficiency, they also present significant risks, particularly in legal contexts. Common risks include hallucinations, outdated or inaccurate legislation, case law, or legal principles, and the potential for misinterpretation of complex legal concepts that require professional judgment.

Because GenAI can produce responses that appear fluent and authoritative, even when they are inaccurate or incomplete, users may be misled into believing the output is correct. This creates the risk of misinformation and over-reliance on GenAI output, which may lead to misunderstanding of legal rights or taking the wrong course of action, which may potentially resulting in costly mistakes, legal exposure, or adverse outcomes.

Users must understand the limitations of GenAI and their responsibilities when using these tools:

They are using an AI tool, not receiving services from a legal professional.
Regulatory protections prohibit non-legal professionals from using AI tools to provide legal services to third parties.
AI outputs are based on model design and training data, not professional legal judgment.
Users accept the risks when choosing to rely on AI instead of seeking qualified legal advice.

Challenge

Create an interactive solution that allows users to safely experiment with GenAI for legal tasks, while learning and practicing responsible use through a “learn as you use” approach. The solution should meet users at where they are in their AI journey and provide hands‑on learning experiences on both the capabilities and limitations of AI in legal contexts, reinforcing user accountability.

The solution should minimally:

Provide a safe environment for users to explore AI-assisted legal tasks
Offer real-time coaching and feedback as users interact with AI tools, such as through contextual nudges, warnings, and explanations that teach responsible use and highlight legal risks (e.g. “Verify authority”, “Possible client confidential information detected – consider redaction”)
Educate users about AI limitations and their responsibilities in an interactive, user-friendly way
Communicate user responsibilities clearly, including disclaimers such as “This is not legal advice” and to seek professional advice.
Incorporate basic safeguards (such as those suggested in IMDA’s AI Governance Framework) to protect sensitive information and reduce the risk of misleading or unsupported outputs.

Reading Materials

Possible Features

Browser extension or plugin which analyses user prompts on AI platforms in real time, flags issues, and suggests better structured prompts with contextual educational tips.
Browser extension or plug-in which analyses AI-generated legal content and provides users with comprehensive risk ratings and explanations. The tool automatically assesses AI outputs on legal platforms, flagging potential issues including factual inaccuracies, outdated legal references, or overly definitive statements, with colour-coded risk ratings (e.g. potential of hallucination) and clear guidance on next steps such as fact-checking requirements or when to consult a qualified lawyer.
Redaction tool that detects potentially confidential information (such as personal identifiable data)before text is sent to a model, suggests or even automates redactions, blocks transmission until user confirmation, and maintains an audit log.
A web or mobile application that provides structured learning modules (e.g. interactive tutorials) where users practice common legal AI tasks using pre-built scenarios and sample documents. Users work through guided exercises such as drafting simple legal letters, researching basic legal concepts, or reviewing contract clauses, with the app providing real-time feedback on their prompt quality and flagging potentially problematic approaches.

Track Prize

$50 Grab voucher per participant for the winning team.
Opportunity to present winning prototype to Ministry of Law, including senior management.

About the Sponsor

Background

The Need for Better Public Discourse

There is a growing need for tools to make public conversations more fact‑based, respectful, and informed. People often struggle to access accurate information quickly, especially during complex social or political discussions, which can contribute to polarising debates and the spread of misinformation.

Parliamentary debates shape laws and policies, but verifying what was said previously – across Hansard, official press releases, MPs’ public statements and so on, remains slow and manual. Inaccurate or out-of-context quotes can distort parliamentary intent, mislead legal research, and weaken policy arguments.

Legal professionals and researchers analyse legislative history and intent to interpret statutes. Today, this often means trawling through lengthy records and media archives, a process that can be time‑consuming and error-prone.

Furthermore, public discourse is increasingly driven by fast‑moving information flows. In today’s environment where misinformation spreads quickly and conversations can become polarised, the ability to surface accurate, verifiable information with proper context is essential to maintain trust and support informed discussions in parliamentary debates, policy dialogues, and public forums.

While publicly available AI tools can help, they have limitations:

Hallucinations: AI may generate plausible but incorrect content.
Knowledge gaps: Models may not include Hansard or the latest statements.
Context risk: Summaries can omit nuances or lack understanding of Singapore’s context.

These factors highlight the importance for solutions that are fast yet reliable, prioritising accuracy, transparency, and contextual integrity to support robust parliamentary debate, legal research workflows, and ultimately to foster healthier public conversations.

Challenge

Develop a solution that enables users to quickly and reliably find out what has been said in Parliament, verify statements against official records, and cross-check information across different sources and timeframes.

Your tool should help users get clear, accurate answers about parliamentary discussions, ministerial statements, policy debates, press releases, and official social media communications while maintaining transparency about sources. The solution should deliver trustworthy information at the speed required for modern discourse, supporting informed parliamentary debates, efficient legal research, and constructive public conversations grounded in verified, credible information.

Possible Features

Advanced search and retrieval capabilities that allow users to query across Hansard records, ministerial press releases, official social media posts, and government websites using natural language questions like "What did the Minister say about housing policy in 2024?"
Source attribution that displays exact quotes with direct links to original sources.
Cross-verification features that identify contradictions across different statements, track policy position changes over time, and provide confidence scores based on source reliability and completeness.
Display information as timeline views for policy evolution, side-by-side comparisons of different statements, or structured summaries with clear source breakdowns.
Live processing capabilities that automatically capture, analyse, and ingest new parliamentary sessions, press releases, and social media posts as they are published, making them immediately searchable and referable with existing records in real-time (or near real time)

Track Prize

$50 Grab voucher per participant for the winning team.
Opportunity to present winning prototype to Ministry of Law, including senior management.