This is a prototype, not a shipped product. I fully designed and developed the system and simulated LLM responses to explore the interaction model. Some use cases connected to real APIs and created real artifacts. It was built to demo at sales, conferences and for investor conversations.
Context
AI tools have made plain English the natural way to interact with products. But enterprise software still expects users to navigate complex workflows in a click-and-discover format. LightBeam has dense, information-rich pages with insights spread across multiple workflows. We set out to see if we could bridge the gaps and allow users to seek information quickly and simply.
Our goals
Reduce time to insight
Let users query across multiple workflows to troubleshoot issues, assess incidents, and triage quickly.
Make users more productive
Surface what needs attention (daily priorities, pending tasks, what’s changed) without dashboard-hopping.
Simplify complex tasks
Guide users through configurations with smart suggestions, reducing reliance on support.
Augment, not replace
Add an intelligence layer on top of the existing UI.
The Problem
A privacy officer gets a call: ‘The company’s Google Drive account is breached. What is potentially at risk?’
To answer, they have to:
- check 5-6 different screens (each specialized and focussed on a specific aspect of privacy)
- export 2-3 CSVs (to gather supporting evidence and validate data)
- spend a few hours correlating data manually
My Approach
I started by creating a comprehensive list of potential questions that users may ask the AI assistant.
Not all queries are equal. Some are one-offs (What is access review?), some are daily rituals (what should I focus on today?), and some are emergencies (we’ve been breached). AI responses need to be different for each.
Most AIs answer in lengthy text blocks or add a few graphs. I wanted responses designed around UX patterns.
Text first is faster and less overwhelming. Rely on this whenever possible.
Examples of a few components that make up the AI chat interface
The hardest queries are the most valuable. “What’s wrong with Google Drive?” required visiting 5-6 screens, exporting CSVs, and manually analyzing data. If we could collapse this into 1 response, we’d prove AI’s true value.
Synthesis is the superpower and also the hardest feature to crack. This would take the maximum amount of effort in terms of the LLM’s training.
An example of how a complex query response should ideally look like.
See it in action
Time-critical queries
‘John’s account has been breached. What is the potential risk?’
Allows users to answer time-sensitive questions promptly with appropriate follow-up suggestions / actions.
Guiding users through complex flow
‘I want to archive old files’
Allows users to configure a policy with smart defaults. Without AI, this would involve going through a 4-step wizard to discover data and create the flow.
Increasing user's efficiency
‘What should I focus on today?’
Allows the system to prioritize tasks for the user when they are overwhelmed with data.
Getting the lay of the land
‘How has my environment changed?’
Allows the system to quickly summarize what has been added, deleted, etc. in the environment. Especially useful when customers are dealing with a large number of datasources, millions of files and hundreds of users who are constantly changing.
Outcomes
Response design system
A complete UX pattern library for AI responses including 5-layer answer anatomy, 12 block types, rules for when to show text vs. tables vs. charts. Ensures consistency across every interaction.
Production-ready PRD
Full product spec covering interaction flows, component specs, edge cases. Engineering could pick this up and build.
Demo-ready prototype
Simulated LLM behavior but wired to real APIs. Live data in responses, some actions actually executing (reports, access reviews). Used for conference and sales demos and investor conversations.
Sales enablement package
Demo guide with exact prompts, sequences, and talking points by audience. Sales could demo confidently without design support.