Search
Search is a web application that lets investigative teams in industries like law perform agentic searches and tabular operations on massive document databases using natural language.
As ConductorAI’s founding designer I built key prototypes that demonstrated its value to prospective clients. Users report Search helps them perform investigations 20x faster. I worked closely with our engineers and customer strategists to figure out the layout of the app in general.
Problem
How can we use LLMs to help teams more effectively perform investigations on thousands or hundreds of thousands of documents?
Finding the metaphor
To solve the problem, I wanted to explore two routes. Should our software solution be interface-first or chat-first?
Interface-first
One idea was to take familiar interfaces for users and augment their abilities with semantic search. This meant less time spent learning a new interface and paradigm and more time investigating.
Chat-first
The second was to fully accept the new chat interaction paradigm and allow users total freedom to explore their documents in plain English. This gave them more flexibility at the expense of clarity.
Synthesis
The solution I came up with was to design an agentic chatbot that could interact with the existing document library interface in plain language to take advantage of the strengths and weaknesses of both. Users report performing investigations 20x faster.
