Dobby: Enterprise AI Knowledge Agent
Private, cited answers that run inside a customer’s own environment. No-code setup, preset-first building, and governed publishing so teams can roll AI out safely.
I led end-to-end design for an on-prem enterprise AI knowledge agent, from discovery to shipped flows.
Non-technical owners stalled at setup, so I made presets the default and moved approvals into a focused publishing lane with audit and rollback.
We shipped fast using AI-assisted development workflows. UI polish is planned for upcoming releases as the product matures.
Industry
Timeline
Responsibility
Team Size
1 Product Designer • 1 Product Manager • 3 Developers
Constraints
Paid users were queued, small team, speed mattered. We used AI-assisted workflows to ship quickly, with UI polish scheduled for later releases.
What is Dobby.ai?
Dobby helps teams build a knowledge base from internal files, so users can get reliable, source-backed answers while keeping data inside the customer’s environment.
For who?
Non-technical admins and cross-functional teams who need AI inside regulated or high-complexity workflows.
What data it uses?
Internal documentation and operational files like SOPs, reports, and log files.
Why on-prem?
Dobby runs inside the customer’s own servers so sensitive data stays in their network.
Problem
Enterprise teams want AI in their workflow, but setup is often handled by non-technical admins, and it can feel risky when the UI is technical and support is limited. Many teams also require on-prem or offline deployment for sensitive data, while still expecting fast, source-backed answers they can verify.
So…how do we make enterprise AI adoption trustworthy, auditable, and secure?
What we learned
From competitors
From potential users
5 user interviews to understand roles, anxieties, and what “trust” means in practice.
From user journey
Journey map helps us to locate the highest-stress moments across build, use, and publish.
Key highlights
Setup anxiety is the real drop-off
People want a guided path that feels hard to mess up.
Trust means “show me the sources.”
Sources beat confidence scores.
Publishing needs guardrails
Approvals, audit trail, and rollback are part of the UX, not an admin afterthought.
End-to-end user flow by role
This diagram maps the full product flow and shows how Admins, Editors, and Viewers move through it, what each role can access, and where handoffs happen.
Solution
Get to a first cited answer fast
Upload → pick a preset → see readiness → ask → get an answer with sources.
Approval you can skim and ship
Request → review → approve → publish, with audit and rollback.
Roles & scoped visibility
Admin publishes, Editor prepares, Viewer chats, with scoped visibility so non-admins only see what they should.
Design System
I built a modular design system so the team could scale faster on upcoming launches with reusable components.
Testing and iteration
We tested with 5 users, then iterated to reduce “where am I?” confusion, clarify readiness states, and make failure recovery faster.
Project Dropdown and Expandable Conversation List
Model Loading State
Error Log Download Button
Impact
Within 30 days of launch, Dobby.ai helped users reduce daily log file checks from 30 minutes to 16–24 seconds, reaching 99% accuracy (vs industry benchmark ~80%)
Setup completion
≥60% create an assistant + send first test prompt in one session
Time to first cited answer
≥50% get a sourced answer within 10 minutes
Final Design & Up Next
Dobby is a web-based, desktop-first product today. We prioritized shipping quickly for active enterprise customers, so the current release focuses on preset-first setup, source-backed answers, and a governed publish lane. Mobile is not deployed yet.
Up next: UI refinements and mobile version