Merlin: AI-augmented Tiered Intelligence Reporting
UX Research
Enterprise Design
Human–AI Teaming
Team members
Sean Ren
Tushita Kaul
Client
The Laboratory of
Analytical Sciences
Role
Research, UX, and Product Design
Project Overview
In this NSA-sponsored project, we partnered with the Laboratory for Analytic Sciences to reimagine tiered reporting, the process of tailoring intelligence to audiences based on their needs, expertise, and security levels. Our goal was to design an AI-augmented platform that utilizes AI to generate versions for multiple classification tiers, while keeping the analyst as an active participant in the process.
Year
2026 (11 weeks)
Research Question
How might the design of an interface automate the tiered reporting process, so that reporters might more efficiently and knowledgeably team with AI to sanitize and deliver sensitive information across classified tiers?
What might the future of AI-human teaming look like in the future?
Rather than asking which analyst tasks a machine could take over, we asked what new work analysts could do with AI as a partner.
Figure adapted from Brynjolfsson (2022), The Turing Trap.
Project Launch
We began the project by meeting with intelligence analysts, where we were introduced to the sanitization and tiered reporting process.
Key Takeaways:
-
Knowing what can and cannot be included at certain classifications is an ambiguous process, one strengthened by experience.
-
Checking content against numerous policies is tedious and time-consuming, which makes it a strong opportunity for AI-human teaming.
-
Different analysts and organizations all structure the reporting process differently, so tools must adapt to varied workflows.
Aisha: Senior Report Releaser
About
Aisha is a Senior Report Releaser with eight years of intelligence experience. She serves as the ultimate checkpoint, providing a "final sanity check" to ensure reports are accurate, properly classified, and meet the intelligence needs of customers.
Responsibilities
She is responsible for final release authorization, which involves meticulously verifying classification markings, reviewing source protection measures, and tailoring distribution channels to specific customer requirements.
User Journey Map
First Sketches
Concept 1:
A side-by-side comparison screen where AI compares two tiers, and surfaces where meaning or content has shifted.
Concept 2:
A node-based "report web" where analysts can see incoming reports, rearrange them based on priority, and
manage needed communication.
Concept 3:
This concept groups related reports under one initiative, and AI connects them so relevant policy and standards appear in context.
First-Round Interviews
Using our initial sketches, we received feedback and clarified aspects of the reporter workflow.
Key Takeaway:
-
Report releasers are recipient focused and will always choose accuracy over efficiency.
Building the Proposed Task Analysis
Key:
Low Fidelity Mockups
Using our proposed task analysis, we broke the dashboard into three segments and began creating software mockups
Mid Fidelity Mockups
Each team member brought their variations into
mid-fidelity to share with the analysts.
Second Round Interviews
Using our mid-fidelity mockups, we presented our ideas to analysts to gather feedback before moving toward the final prototype.
Key Takeaways:
-
Give reporters choices in how they navigate and review their documents.
-
Allow for control of complexity and progressive disclosure of information.
Customizable Workspace
Reporters can query the AI to create semantic tags that reorganize the workspace around their current focus.
Quick-sort buttons instantly regroup reports by priority, the feature analysts valued most.
Smart Tasks
With one click, the AI generates a task list from the report, each with a quick link that takes the reporter directly to where the work needs to be done.
Reporters can jump straight to each task, allowing them to reorient themselves each day.
Integrated Coordination
Correspondence and organizational approvals live inside the system, so reporters never have to switch tools.
The AI drafts coordination requests that reporters can edit and send, with approval status tracked alongside each report.
Local Node View
The local node view shows every action, edit, and coordination on a report as a timeline.
Reporters can trace a report's full history at a glance, and every change is memorialized for future reference.
Global Node View
The global node view zooms out to show how a report connects across the wider reporting initiative.
Reporters can surface overlapping intelligence, customers, and policies, helping to find connections before release.
Two Monitor Workspace
From our research, we found that a two-monitor workstation was common practice for intelligence reporters.
AI Review Flags
AI flags call out areas for review based on the customers receiving the report, so each tier is tailored to its audience.
The AI can make changes for the reporter, but reporter reasoning is required, creating a record and future training data.
Multi-tier View
Tiers display side by side with content differences highlighted. Hovering over a highlight in one tier lights up the matching content in the other, making it easy to move between tiers.
AI Tier Comparison
By querying the AI, the tiers will auto-scroll to related content across tiers. In the chat, the AI summarizes how the wording was sanitized at each level.
Can be utilized in comparing distinvtive reports.
Margin Edits
When a report returns, annotations and comments from other editors are memorialized in the margin for the releaser to confirm.
Each change is also visualized in the document, making review faster.
Final Checklist
The final checklist is the releaser's last confirmation before dissemination.
They can verify that all suggested edits and customer requirements have been satisfied before releasing.
Final Thoughts
This project was a valuable opportunity to work alongside stakeholders, centering their needs and pain points to design a solution that serves not one analyst, but many. Feedback from LAS showed that every releaser works differently, which pushed us toward a customizable workspace that adapts to each user rather than forcing a single workflow. Merlin also challenged me to imagine the future of AI-human teaming, and to consider how UX design can support human judgment rather than replace it, keeping the releaser accountable for every decision. Moving forward, I want to push my ideas even further and explore how AI can transform the reporting workflow itself, not just speed up the one that exists today.