From developer MVP to pro-level settings
Founded in early 2024 by YC W23 alums Rishabh Srivastava and Medha Basu, Defog's first build was an AI data analyst for enterprises. The product was based on SQLCoder-70B, a best-in-market, open source, fine-tuned LLM for text to sql generation, created by the company. Defog was originally designed as a powerful data analysis tool with an array of use cases across many industries. You can learn more about them here.
The aim of this project was to take the clean cut product created at an early stage and make it market-ready, implementing a solid brand language, improved usability and seamless user experience for a more market-ready product. New features such as the Defog Oracle were added to generate rich reports and analyses, while maintaining the customisability of the model and source data.
It Starts with Information Design
The product's initial user flow was essentially a step-by-step process to calibrate the model and data source for the user's needs. The actual AI data analysis assistant was hidden behind this, which largely impaired usability to those who intend to only ask questions and get analyses without going through the entire process of setting up and aligning the model, conducting checks, and troubleshooting errors. Error management was relatively slim, without appropriately timed notifications. The process of calibrating the model and data source, checking errors and setting up the tool was almost impossible for an everyday users owing to a lack of visual cues and affordances. The user journey did not account for the most user outcomes and motives, and did not appropriately prioritise the ones it did account for.
My first intervention was to propose a new information architecture that separated user outcomes for the two types of users- Admin users would need to focus on setting up the back end of the product to function properly for their business/use case, and Team/End users would be asking questions, generating reports, viewing and toggling between available databases, editing basic settings and communicating with their team within the product environment. This new structure allowed us to create an experience that suited the needs of all users while highlighting the right product capability at the right moment and allowing smooth usability for both individuals and enterprise teams.
The next step involved fleshing out the various user flows for admin users, team users, and their collaborations. This highlighted edge cases that ironed out hiccups in the experience and provided myself as well as the Defog team with a blueprint and scope for the product overhaul ahead. What followed was weeks to brainstorming, iteration and research to arrive at the best possible final output in every flow and screen, in terms of visual impact, brand recognition and silky smooth product experience.
Led by Research
This was my first time working on an AI-enabled product, and every step was underscored by rigorous research to make sure I was adhering to the highest standard in an industry where the cutting edge is… Us! In an industry where the technology is developing at breakneck speed, everyone is trying something new everyday, and UX rules are still malleable, I had an incredible opportunity to learn on the job, do my own research and theorise best possible design strategies. This is a design blackbox I'm only just beginning to play with, and I wrote about it in my article about AI-enabled products.
Defog is not currently live as the company is in the process of cooking even more open source goodness as AI surges ahead and unlocks incredible possibilities. Keep up with the founders here!
