Team Info
Team 10 - Jamalware
Inspiration
When we considered problem 4, we could understand the challenges legal laymen may face when opting to be self represented in legal actions, such as a complicated process, daunting legal jargon etc. As such, we sought to use AI in a way that would help users navigate this rather complicated landscape. We thought it would be best if AI could be used in a way where it was a helpful assistant that focuses on expressing ideas the user already had, rather than give inaccurate legal advice that would potentially trouble users even further.
What it does
Our project is able to help users create a coherent expression of the case facts in their specific situation, assist them in understanding the pre-filing and filing process, as well as prepare them for hearings by assisting with case chronology and making their arguments and facts focused.
How we built it
We divided the responsibilities for the project. For the technical members of the team, one member focused on developing the user interface, while two members worked on building the backend and integrating the backend functionality with the UI. The legal members focused on understanding the relevant legal processes, as well as testing hypothetical scenarios on the test prompts and assessing the output to identify areas for improvement. Both the technical and legal sides worked closely with each other throughout the project's development.
Challenges we ran into
Firstly, the website for the Small Claims Tribunal portal underwent maintenance right after lunch on the first day of the hackathon, which made using it for reference very difficult. Fortunately, we were able to get around it by working on our project as a separate webpage, rather than focusing on how it could be integrated into the portal immediately.
We also found prompt engineering to be quite a prominent issue towards the end of the project. In the spirit of responsible genAI usage, we did not want the output of our AI to lean too heavily towards “legal advice”, as such it was an interesting challenge to balance the quality of the output with ensuring that the AI stays in its role of simply refining what the user already had in mind.
Accomplishments that we're proud of
We think that we have managed to create an app that is functional and user-friendly given the timeframe. We also think that our project has potential and can be taken further given more time and resources. We think that through this project, all of our members have learnt something about our own as well as another discipline of study. We think that we have also managed to maintain the fine balance between responsible usage and abuse of AI.
What we learned
We learnt more about the processes involved in making a claim at the Small Claims Tribunal, including all the information that have to be submitted upon filing a claim. We previously didn't understand how complicated it may be for a self-represented person to file a claim. As such, we also learnt the importance of designing around the actual needs of the users.
We (especially the legal team) gained a better understanding of the limitations of LLMs and the importance of designing clear, structured prompts. We learnt how to refine prompts by testing the system with different factual scenarios and identify areas for improvement. In particular, we found that the way information is structured and the instructions given to the LLM can significantly affect the quality of its output.
The technical team also had the opportunity to explore other parts of website development such as UI which they previously may have been unfamiliar with.
What's next for Small Claims Tribunal Interactive Guide
We have considered several potential extensions that could be developed if we had more time. The next stage of development would be to help users organise the facts of their dispute into one coherent and persuasive argument. The system could also identify potential counterarguments and weaknesses in the user’s position, allowing them to better prepare for the other party’s response. This would extend the project beyond preparing the facts of a claim and provide support during the e-Mediation or hearing stage, which can be particularly challenging for self-represented persons.
Another potential extension would be to use an LLM trained or fine-tuned on appropriate production data to improve its ability to reason about disputes in the context of the actual tribunal process. This could potentially make the system more accurate and better adapted to the types of factual issues and arguments that arise in real cases.
Log in or sign up for Devpost to join the conversation.