Wisetax GPT

To make their search more efficient, they integrated a LLM (Large Language Model) into their search engine to allow their users to conduct natural language searches. However, they encountered design difficulties due to the dichotomy between their traditional keyword search and natural language search.

Brief

Wisetax contacted me at the end of 2023 to conduct a small UX audit of possible improvements to their platform and to design the integration of their new natural language search functionality.

Challenges

• Allowing contextual conversation, not just a search that returns a single unique result.

• Understanding the core of a tax attorney's work: when building a case, the most important factor is the relevance of the sources used. Thus, the UX must emphasize the source documents that form the basis of the LLM's proposed response.

• Implementing natural language search while still allowing users to conduct advanced "classic" searches by keyword, sorting, and filtering within the legal corpus.

• Proposing a simple reorganization of the interface following the audit to best highlight the new search journeys.

• Ensuring the easiest possible integration of changes, given that the technical team consists of only one backend-focused developer. The front-end was built with Vue.js using a Google Material Design UI. This technical basis was used to create my changes in Figma.

• 10 days' worth of total design budget: challenge accepted!

Design methodology

• UX specifications: Value proposition, personas, identified problems, possible solutions, and design job objectives.

• Quick UX audit: Tour of the application with the CTO to note the different user journeys, known issues, and technical difficulties. This was an opportunity to deeply understand the needs of tax attorneys.

• Wireframes: Following the audit, I created iterations of wireframes based on the most common UX patterns relative to Wisetax's journeys, to clarify usage. Proposals for reorganizing navigation and the journey related to natural language search.

• Moodboard: Moodboard of existing and recognized LLM platforms to focus design efforts on the most challenging part: source review and analysis.

• UI Prototyping: Based on the existing Material Design UI philosophy, proposing a new, cleaner look on Figma, allowing more room for future platform developments.

• Creation of a playable demo: Allowing the developer to understand screen sequences, interactions, loading and error states, and the conversational flow.

Here is how the platform was before:

And how it became after redesign:

Sources relevance:

Sources reviewing:

Some component states:

UI experiment:

Thanks for reading it all!