{{item.label}}
{{item.t1}}{{item.t2}}
{{tb.text}}
← All work
NSA, Laboratory for Analytic Sciences
4 months, team of 4
ComplAI. AI experts can trust.
An AI assistant that helps intelligence analysts stay compliant with laws that keep changing, without taking away the complex work they actually enjoy.
Your turn
What should AI do?
Analysts at NSA's Laboratory for Analytic Sciences spend their days on tasks like these. For each one, decide: hand it to AI, or keep it human? Then see what our research found.
Matched our research{{an.score}} of {{an.done}}
Let AI handle it{{item.t}}
Keep it human{{item.t}}
That's the line we drew: automate the tedious, not the thinking. Analysts love the hard parts.
The task
{{an.task}}
{{an.rt}}
{{an.task}}
What analysts told us · {{an.ansT}}{{an.why}}
Your result
You matched {{an.score}} of 7.
{{an.endT}}
Problem
Laws and policies kept changing, so analysts wrote overly cautious queries to stay safe.
Why
They relied on outdated tradecraft repositories and had no real-time guidance.
What we did
An AI assistant that detects risk, alerts and recommends, inside the analyst's workflow.
Result
Research, wireframes and iterative testing improved clarity by 35% in sensitive decision workflows.
The problem
Rules kept moving. So analysts played it safe, and safe killed creativity.
Client: NSA Laboratory for Analytic SciencesUsers: Target Network Analysts
Laws keep changing.
Fear means cautious queries.
Tradecraft is outdated.
Guiding question
How might AI help analysts work faster and smarter, and stay compliant?
SearchAI-supported queries
VisualizePatterns in the data
LearnTradecraft by doing
Test scenarios
{{item.title}}
{{item.text}}
What we learned
The first interview broke our assumptions.
Tap a card to see what we found.
Who we designed for
Her "proceed with caution" moment
“{{item.q}}”
{{item.text}}
The full journey map. Click to zoom.
My role
UI and UX with a team of four. Research to prototype.
User researchUser flowsPrototypesAlerts and recommendations UIFigma
With Soumya Batra, Leah Tatu and Gabrielle Thorpe
Constraints
ConfidentialNo access to real tools
New to AILearning as we designed
MultitaskingThe tool must earn attention
Sensitive dataPrivacy shapes everything
Key decisions
Five decisions behind ComplAI.
Switch between what we saw, what we did, and what we chose not to do.
Under the hood
Sensitive data in separate lanes.
LLMRAGCognitive SearchAI Search
Each search runs on its own data, so classified sources never mix.
Iterations
From whiteboard to widget.
The concept changed shape four times. The biggest shift came from sitting with analysts and hearing what wouldn't work.
{{item.tab}}
{{item.title}}
{{item.text}}
What changed
{{c}}
The result
Detect. Alert. Recommend.
Knows what you're working onFlags risk as laws changePoints to current tradecraft
How to use ComplAI, a full walkthrough of the prototype.
Key features
Help at the moment it's needed.
Alerts inside the workspace
ComplAI surfaces context and related resources right next to the analyst's work.
{{item.t}}
{{item.d}}
35%improvement in clarity in sensitive decision workflows, through research, wireframes and iterative testing.
What I learned
Good AI design is knowing what to leave in human hands.
One visionMany voices, one design
Stay flexibleGoals evolved as we learned
CommunicateEmpathy keeps teams aligned
Where it could go next
Help form queries, not automate them
Goals that tailor every recommendation
A test mode trained on live policy updates
Process in phases
User interviews
Benchmarking
Whiteboarding
AI idea cards
Audit trail
Audit trail detail