Designing Client-Consultant chat system
Designed a sidebar UI to connect clients & consultant contextually over chat instead of email
Berlin
Team
CTO, CEO, Lead Consultant, +Me
Role
Product Designer
Shipped
2023
Designing MVP for B2B AI Legal Assistant in 2023
Atla AI
B2B AI legal assistant
(YC Summer '23)Remote
Contribution
Helped founders strategise with clarity. Designed end-to-end product with handoff. Designing new AI design patterns at the start of the AI era was a privilege.
Role
Freelance Designer
Team
Co-founders, +Me
Shipped
2023
Overview
Atla uses AI safety techniques to make sense of the dense legal language, serving as a reliable ally in high-stakes legal decision-making. atla searches through millions of legal docs, then uses machine learning to extract key information from each source. This process-based reasoning leads to reliable answers.
This was in 2023, at the dawn of AI, and designing new AI design patterns from scratch for a new tech paradigm was a privilege.
Working setup
I worked with the founders to refine their vision into an actionable, iterative product roadmap. Interviewing real users and doing usability testing helped us leapfrog with the iterations. I was able to come up with solutions that'd solve users' problems while keeping the MVP scope small.
User flow
The users i.e. the legal teams at Volkswagen and N26 ask a legal question, and the AI assistant generates a reliable answer alongside the legal sources, documents and court decisions as citations. The tool lets them ask follow-up questions, and dig deeper to find the 'whole' answer.
Insights
Dig deeper
On interviewing the lawyers, the biggest hurdle in adopting the software came to light. Lawyers needed to reduce liability, and a one-shot AI-generated summary did not cut it for them. They needed a way to go deeper into their research and find answers from different angles. After testing a few prototypes, we went with a threaded way of asking for follow-ups - Dig deeper.
Increase credibility
Other major insight was a need to instil trust in the answers. In 2023, AI was not a proven technology to do deep research, similar to how it is now. Adding to it, that the users were not the most tech-savvy demographic, we needed to find ways to help them trust the summaries and answers given by the tool.
Avoid tool switching
As much of the lawyers' work was done through email, it became evident that getting data from email threads into atla's tool required a lot of manual effort and was error-prone. The users had to summarize everything from the email into short prompts, creating subpar results. As an initative, I proposed a Gmail add-on to summarise the email thread and start a conversation with atla right from Gmail.
Design logistics
At pre-seed state, I wanted to reduce any design-dev handoff barriers. To enable smoother DX, I gave clear handoff documents so that any design confusion does not break CTO's (only dev) flow.
Outcome: Summer 2023
Roman and Maurice did a successful YC launch in July 2023 and the MVP was being used by the legal team of N26 and Volkswagen. Later down the road, atla decided to pivot into its current model.
Pivot: Fall 2023
In the fall of 2023, decided to pivot to a deep AI model into an evaluation platform for AI agents. Here are some of the designs for the MVP.
Next project