- Client: DFDL Cambodia.
- Role: Operations Efficiency Consultant (Director's Office)
- Duration: 6 Months
Project Overview
The special project aimed to create an AI-powered chatbot using OpenAI's language model, integrated with Nuxt.js and Tailwind CSS, to serve as an intelligent assistant for DFDL advisors.
The chatbot was designed to handle queries related to the Cambodian Law on Investment, the Cambodian Law on Commercial Enterprise, and the Cambodian Law on Taxation, facilitating faster and more efficient legal research and advisory.
Objectives
- Enhanced Efficiency: Provide DFDL advisors with a quick and reliable tool for accessing specific legal information.
- AI-Driven Insights: Implement a chatbot powered by OpenAI to offer accurate and contextually relevant answers.
- User-Friendly Interface: Develop an intuitive and responsive interface using Nuxt.js and Tailwind CSS.
Development Process
The development of the AI knowledgebase was initiated as a means of improving the speed and efficiency of advisory work within DFDL. I saw the opportunity to develop a dedicated, internal tool that could be scaled with more resources and legal documentation in Cambodia, providing an edge in an increasingly competitive advisory and legal market.
To ensure a quick proof-of-concept, I opted to develop a chatbot solution using Nuxt.js, Vue.js, Tailwind CSS and the OpenAI API which enables natural language processing and understanding.
Delivering a Strategic Edge
The strategic advantage of this project stems not only from its exceptional user experience but also from the curated, custom dataset comprising the Cambodian Law on Investment, the Cambodian Law on Commercial Enterprise, and the Cambodian Law on Taxation.
By concentrating exclusively on this specific set of legal data, as opposed to broader models like ChatGPT that access the entire internet, the chatbot delivers results that are significantly more focused, relevant, and useful to the end-users—in this case, the advisors at DFDL Cambodia.
Development Process and Technology
The development stack for this project needed to be quick, reliable and lean so that a proof of concept could be showcased at a quarterly board meeting.
Design and Architecture
- Chose Nuxt.js for its robust server-side rendering capabilities and ease of integration with Vue.js.
- Selected Tailwind CSS to ensure a modern and responsive design.
- Integrated OpenAI’s API to power the chatbot, enabling natural language processing and understanding.
Deployment and Training
- Deployed the application on Fly.io, a reliable and scalable cloud platform.
- Provided comprehensive training sessions for DFDL advisors to ensure smooth adoption and effective usage of the chatbot.
The various instances of this are available to test online via the following links:
- Law on Investment Bot
- Law on Commercial Enterprise Bot
- Law on Taxation Bot
- Tax Audit and Crime Investigation Bot
Challenges and Solutions
Whilst the data was readily available online, the formats and complexity of the material varied significantly between each ministry. As a result, the first step in the process was to create a custom scripts that would take the raw data and standardize the dataset to ensure consistency before integrating with the application. The script developed is open-source and available on my GitHub.
Alongside this, it was essential to fine-tune the OpenAI model to understand legal terminology and context-specific queries. This required iterative testing to achieve the desired level of accuracy, particulalry as the dataset was specific to a relatively small niche (Cambodian corporate, commercial and tax law in this case).
Finally, it was critical to ensure that DFDL advisors were comfortable using the new tool involved. So it was essential to deploy human-centric design techniques of interviewing, training and gathering feedback to facilitate smooth adoption.
To find out more about how I use human-centric design principles to master problem solving, consider downloading my free ebook.
Future Enhancements and Opportunities
Looking ahead, several enhancements are planned to further improve this proof of concept. One key enhancement is expanding the dataset to include additional legal documents and resources, making the tool even more valuable by providing a more comprehensive repository of legal information. Another planned enhancement is incorporating multilingual support to cater to non-English speaking advisors and clients, thus expanding the tool's accessibility and usability.
More significantly whilst this tool was designed for internal use only, I believe it offers the potential to be a pivotal moment in the legal landscape in Cambodia as the tool that not only enhances the efficiency and effectiveness of DFDL advisors, but also sets a new standard for internal legal research and advisory.
Whilst the DFDL specific version of the offering remains for internal use only, you can see a demo of the exact same approach via an app I called BuffetBot.