| ID | Description | Priority | State | Contributors |
|---|---|---|---|---|
| 1 | Integrate with vector database for data retrieval | MUST | ✔ | Darius & Zhaoqi |
| 2 | Understand users input and transfer into multiple tasks | MUST | ✔ | Team |
| 3 | Use LLM for interactive chat with users | MUST | ✔ | Darius |
| 4 | Answer specific questions based on its knowledge base | MUST | ✔ | Team |
| 5 | Generate, store, and manage chat logs | MUST | ✔ | Faisal & Zhaoqi |
| 7 | Convert PDF documents to text for analysis | MUST | ✔ | Faisal |
| 8 | Used through a web browser | MUST | ✔ | Zhaoqi |
| 9 | Ability to summarise a document | SHOULD | ✔ | Darius & Zhaoqi |
| 10 | Extracting data from documents | SHOULD | ✔ | Faisal |
| 11 | Provide accurate legal information and precedents | SHOULD | ✔ | Darius |
| 12 | Give suggestions for improvement on user input files | SHOULD | ✔ | Team |
| 13 | Translate foreign documents | COULD | ✔ | Zhaoqi |
| 14 | Create template documents | COULD | ||
| 15 | Implement AI to predict case outcomes | WON'T | ||
| 16 | Ability to create arguments, opinions to be used in court | WON'T | ||
| Key Functionalities: | 100 % | |||
| Optional Functionalities: | 83 % | |||
| Work Packages | Zhaoqi | Darius | Faisal |
|---|---|---|---|
| Partners liaison | 20% | 30% | 50% |
| Requirement analysis | 33% | 33% | 33% |
| Research and Experiments | 25% | 50% | 25% |
| Coding | 33% | 33% | 33% |
| Testing | 40% | 40% | 20% |
| Report Website | 60% | 20% | 20% |
| Video Editing | 35% | 25% | 40% |
| Overall Contribution | 33% | 33% | 33% |
During the development, we always remember to put users at the center and always trying to improve user experience.
• We have a clear linear workflow that obey a usual workflow: Raise question - Separate it from a large one to multiple smaller ones - Solve question.
• There are always warning message when an error is raised from users' mistakes during the usage.
• We have a clear design for user to find the main Functionality quickly.
• We have also made the main process re-doable so that users can always regenerate the response which gives them some freedom.
The main functionality we focused on is the development of an AI agents system.
Meanwhile we have fufilled most of the client requirements and the MoSCoW list. Finally we came out with a working system using the concept of AI agents that could satisfy most queries and analysis in legal area specific to a certain database. And the process is splitted by tasks working with different agents.
The stability is mainly based on the APIs we are using. For now we are using Azure Open AI API keys, which are provided by Azure and it provide a stable service.
The frequency, as same as the stability, is mainly based on the APIs we are using. However due to the budget and usage limit we are given by the client, we can only using gpt-3.5 model and even that the frequency of multiple chats would still be limited. However, it should be enough for normal usage.
JurisBUD AI is deployed using docker service. So the whole project is dockerized and easy to be visited straightly. It is also benificial for deployment if the website version using DNS is needed in the future. Meanwhile, the database for the system is able to be customized using legal database from any clients/users that need the service, it could also improve the data privacy and confidentiality, which are the most important aspects when we are dealing with leagl documents.
The project is modularized well. The key functionality is the AI system and all features are developed separately with the integration at the end. This means that if an error is reported we can fix it just with the specific module but the rest of the application would not be affected.
We did a good job in project management. The project is separated into 3 main parts -- Frontend, Backend and the AI System. Each of us in the team are responsible to one of them. We utilized gannt chart to schedule the timeline and we have our own deadline for each parts which bring us extra time for integration and debugging. Meanwhile, we also have a good communication with the client who supported us a lot and brought us some amazing ideas.
| ID | BUG Description | Priority |
|---|---|---|
| 1 | External login is not available | Low |
| 2 | Production docker does not build | Medium |
| 3 | Django URL warnings | Low | 4 | Various nextjs http request errors, doesn't affect experience | Low |
1. Better customization for multiple tasks:
Due to the time limitation and the cost of extra APIs, we are not able to customize specific task in the task list for now. We would likt to have it for more freedom to our users.
2. Multilanguage support:
In the world of legal industry, English is the most commonly used but not the only language written in documents. Communication matters in this area. We will look into the feature about analysing documents from other languages and give a relative response.
3. Website deployment:
Since the database we used are highly private to the specific group of users, we have not implemented it on a web server with empty database -- which might provide less applications. In the future, we are considering to have it online with some open-sourced datasets that could be used in public. It could help more people to solve specific tasks especially if some of the public users are intrested in some open-sourced cases.