SALI LMSS search tool: Automating SALI tagging with 99.1%-accurate AI search results

Project overview
When law firms and their corporate clients start discussing billing, it may take them a while to agree on a common taxonomy. That’s part of a larger challenge in the world of legal services: the lack of an interoperable, universal taxonomy for all legal definitions.
The SALI Alliance decided to tackle this issue with its Legal Matter Standard Specification (LMSS). Its second version, released in 2022, unifies taxonomy and includes billing definitions.
As great as the idea of the SALI LMSS is, however, it’s not exactly user-friendly for either law firms or their corporate clients. Even with the SALI LMSS database at one’s disposal, mapping all definitions to their appropriate SALI tags can take hours upon hours of manual work.
Our client, 273 Ventures, noticed, and that’s how the idea for its SALI LMSS search tool was born. Using the power of generative AI and natural language processing, this tool would automatically suggest relevant SALI tags for the entered legal texts or queries.
However, to make it a reality, 273 Ventures needed a partner with experience in AI models, full-stack web development, and human-centered design. Darly Solutions fit the bill.
Services


Challenges
To implement the web app’s natural language processing (NLP) capabilities, the client wanted to rely on an existing AI model to capitalize on time-tested accuracy and processing capabilities. That’s why, after some deliberation with the client, we opted to integrate the tool with OpenAI’s ChatGPT via an API.
This integration had to perform well and respond promptly while transmitting data securely to and from OpenAI’s servers. Therefore, we had to conduct extensive, thorough API testing.
Finally, 273 Ventures wanted the tool to be a valuable productivity enhancer for its intended target audience. So, we had to ensure its user flows met their expectations and needs. That involved minimizing response time and adding the CSV export feature, among others.
During our discovery phase, we identified the project’s technical requirements and their underlying business needs:
Strategic business needs
01 Launch a SALI LMSS search tool for law firms and their clients
02 Make it available to any visitor without registration or authentication
03 Ensure users can securely and easily process large texts or sensitive data via API or source code use
04 Achieve a user satisfaction rate above 85% for the tool during user testing
05 Improve brand recognition by making the tool free to use
Technical requirements
01 Develop the frontend and backend for the SALI LMSS search web app
02 Integrate the tool with ChatGPT for natural language processing of the search queries
03 Ensure the accuracy rate for the search results above 95%
04 Keep the application error rate below 10%
05 Integrate the tool with the SALI LMSS database via API
06 Enable users to export results in the CSV format
07 Attain a task success rate above 90% for the ensemble of the tool’s features
08 Develop the tool’s API to allow users to process sensitive data or large texts
09 Minimize the response time for producing search results
We’ll build a convenient-to-use tool that leverages advanced AI models like OpenAI’s GPT-4 while ensuring their accuracy and reliability.


.webp)
Solutions
To build a dynamic and fast interface for the tool, we chose React as the main frontend technology. Its component-based architecture was perfect for making the codebase lightweight, ensuring high performance as a result. The backend was powered by Node.js and Next.js, two JavaScript frameworks known for their powerful capabilities for building scalable, fast web apps.
We delivered a fast, dynamic online SALI LMSS search tool that quickly returns the most likely SALI tags for the entered legal texts with an accuracy rate of 99.1%. Users don’t need to create an account or log into one to take advantage of its features.
With the tool’s help, users no longer have to spend hours manually identifying relevant SALI tags. AI-powered search results can now do it in a matter of seconds.



Search bar
Right below the basic information about the tool sits the search bar where users can enter the legal text they need labeled. They can also enter a custom query, instead.
Search results are generated and updated seamlessly as the user enters or tweaks the input, without requiring any action on the user’s part. Users can also include or exclude specific words from the query or specify its beginning and end.
We tested the search workflow not just for accuracy and performance but for usability and intuitiveness as well. Based on the testing results and the subsequent tweaks to the search bar, the task success rate reached 92%.

ChatGPT and SALI integrations
We integrated the tool with ChatGPT via REST API to enable state-of-the-art natural language processing for analyzing its input. Thanks to continuous sync with SALI’s GitHub repository, the tool always uses up-to-date results. We thoroughly tested the accuracy of the AI model's results and optimized performance to minimize response time.
Our improvements enabled the tool to generate responses in under a second, on average. We also tested the AI model’s reliability, which led to the final accuracy rate of 99.1% after our fine-tuning.

Search results
By default, the generated search results appear in a table view, providing a quick overview of relevant tags. The table contains columns for the identified narratives, sections, labels, matches, and definitions, as well as the information sources and reliability scores.
We streamlined and simplified the table view for search results, making it easy to review and browse. We also ensured that the table was dynamically updated as quickly as possible based on changes in the search query.

CSV export
During user research, the ability to easily export search results was frequently mentioned as a desired feature. Based on that research, the CSV file format emerged as the most suitable export option.
So, we implemented the CSV export, making it simple and fast in the process. The tool doesn’t require any extra information for the download to begin: it literally takes a single click to download the search results. What’s more, users can choose which results to include in the table.

API and source code use
In some cases, users may be hesitant to share their legal texts with an online tool because those contain sensitive data. In others, they may need the tool to deal with large text volumes that an online tool may struggle to process.
In either case, our client wanted to meet users halfway by enabling them to use the search tool via an API or run the source code directly. We implemented support for both options while ensuring they remained secure and performed well.
Tech stack
Impact
As a result, 273 Ventures reaped a number of benefits with our performance-optimized, AI-powered SALI LMSS search tool, such as:
We’ll help you capitalize on existing AI models while adapting them to match your specific use cases.


Related case study
Connect with us
.webp)
We are a tech partner that delivers ingenious digital solutions, engineering and vertical services for industry leaders powered by vetted talents.