A hackathon wireframe that became an enterprise Machine-Learning security product, shipped to Fortune 500 customers on a shoestring budget.
The stepping stones toward AI-readiness: designing for machine learning before the industry had a name for it.
Before · rule tables, IP lists, spreadsheets of intent
After · the network becomes a conversation
01 / TENSIONThe problem
Security at scale is impossible to manage manually.
MY ROLEFramed the problem with the NSX PM & security BU lead; translated “too many rules” into a concrete UX thesis.
In a large enterprise datacenter, tens of thousands of workloads chat with each other every second. Security engineers were expected to write firewall rules for each conversation: by hand, by guess, often by ticket. When something broke, no one knew why. When nothing broke, no one knew if they were safe.
VMware had the data. Every flow, every packet, every policy. What it didn't have was a way for a human to see it and act on it without drowning.
2 VMs
1 Link
Simple. One conversation, one rule. A human can hold this in their head.
4 VMs
6 Links
Manageable. A few policies cover it. Still feels like a spreadsheet problem.
12K+ VMs
Millions of Flows
Chaos. No human can hold this. The security model breaks, and nobody notices until 2 AM.
Fortune 500 network admins at 2 AM. They need answers that are fast, clear, and trustworthy. Not more dashboards.
02 / SPARKThe hackathon
48 hours to prove it.
MY ROLESolo designer: sketch, wireframes, pitch deck, demo walkthrough. Shipped as a working narrative in two days.
On a flight back from LAX, I sketched a different answer on an airplane napkin: what if the machine did the hard seeing, and the human only had to confirm? The topology became the interface. Recommendations, not rules. Confidence, not certainty.
How can we ensure users visualize their complex network in action and remediate gaps in a smart, effective way?
Two days later, the wireframes were a working pitch. A quarter later, we were on the VMworld keynote stage.
1
Sales Engineer
Customer pain, field intelligence
1
Product Manager
Roadmap, prioritization, scope
1
Design Manager
Executive air cover, strategy
ME
Designer
Research, workshops, prototypes, validation
LAX → SJC · the napkin
VMworld · on the stage
VMware hackathon · 2018
✓
Output
Hi-fi design proposal shipped in 48 hours
✓
Result
Unit head greenlit the project. Codename: PACE.
6
Months to VMworld
From hackathon win to keynote stage
03 / PROCESSThe journey
The process: a rigorous, messy journey.
MY ROLEDrove the end-to-end process: research, ideation workshops, prototyping, and customer testing at VMworld.
The official process had four clean stages. The real one had loops, dead ends, knowledge gaps, and rework. That's what real product design looks like. The clean arrows come after.
01
Research
Requirements, user flows, competitor analysis, architecture review
Paper prototypes to wireframes to high-fidelity mockups
04
Test
VMworld event testing, synthesis, iteration
Knowledge gaps, recruiting delays, failed tests, shifting briefs. That's what real product design looks like. The clean arrows come after.
04 / CONSTRAINTSDesign challenges
Three constraints that shaped every decision.
MY ROLETranslated raw constraints into design principles; worked with the ML researcher and platform team to define what was possible within each.
01
Abstraction
Challenge: How to show 12,000 VMs without chaos? Approach: Progressive disclosure + hierarchical slicing. The UI has to be smarter than a flat list.
02
Scale
Challenge: Rendering 1.2M network flows in-browser. Approach: D3.js + backend optimization, targeting 200-2000ms. Every millisecond of lag erodes trust.
03
Trust in ML
Challenge: Users skeptical of automated security decisions. Approach: Confidence scores + simulation + manual override. Building AI-readiness starts with earning trust.
05 / ALIGNMENTShared direction
From research findings to shared direction.
MY ROLEFacilitated cross-functional workshops with engineering, PM, sales, and leadership to build shared understanding before a single screen was designed.
Before designing a single screen, I needed every stakeholder aligned on what we learned and where we were going.
01
Synthesize
Distilled interview data, competitor analysis, and user pain points into a clear narrative for the full team.
02
Align
Facilitated workshops to build shared understanding across engineering, product, and sales teams.
03
Co-create
Collaborative design sprints to surface ideas, identify technical constraints, and prioritize what to build first.
04
Validate
Validated assumptions and early concepts directly with enterprise customers before committing to development.
Alignment isn't a meeting. It's a shared artifact. Every stakeholder left with the same mental model of the problem, the constraints, and what "good" looks like.
06 / CRAFTFrom paper to pixels
Four stages of fidelity, each one a different argument.
MY ROLEOwned every stage: wireframes, motion, prototypes, visual. Paired with the ML researcher to make confidence a first-class UI primitive.
Designing with ML meant designing with uncertainty. Every stage had a different job: get the shape right, then the motion, then the trust, then the math.
01
Wireframes
Could a human parse thousands of flows at a glance? Topology-first beat table-first every time.
02
Interaction prototypes
How does filtering feel when your selection changes the recommendation? Responsive, not reactive.
03
Visual polish
Confidence badges, trust affordances, explain-why panels. The ML had to show its work.
Sketches
Wireframes
Hi-fi · dark
Shipped
07 / RESEARCHThree insights
Three findings that shaped every pixel.
MY ROLELed user research with 8 enterprise security admins; synthesized findings into three principles that became the design's non-negotiables.
Trust is earned in failure modes. Admins cared less about the 92% of recommendations that were right and more about what the product did with the 8%.
Recommendations must be reversible. A "one-click apply" with no rollback was dead on arrival in a security context.
The topology is the workspace. Lists are where people leave. Maps are where they stay.
01
Build trust through transparency
Users won't blindly trust AI. Show confidence scores (87%, 92%). Always explain the "why." Explainability was the #1 demand across every interview.
02
Simulation is non-negotiable
Must see network impact before enforcement. One wrong policy could take down the entire network. Every user wanted to see impact before committing.
03
Progressive disclosure, not simplification
Include all functions without overwhelming. Context-aware UI reveals what's needed when. Users consistently wanted to tweak AI suggestions, not just accept them.
08 / SOLUTIONReady for VMworld
NSX Intelligence: the four-screen solution.
MY ROLEDesigned the complete end-to-end experience: topology visualization, discovery wizard, rule review, and simulation flow.
The solution crystallized into four interconnected screens, each targeting a specific moment in the security admin's workflow.
Topology View · ML-powered
Discovery Wizard · scope + config
Review Rules · ML recommendations
Simulate + Publish · safe deployment
09 / VMWORLD3 days of live testing
VMworld 2019: three days, live customers.
MY ROLEDrove the VMworld demo script, keynote visuals, and post-launch UX reviews; translated feedback into the NSX-T bundled experience.
The hackathon prototype became a bundled add-on to the NSX-T platform and the anchor of the VMworld 2019 keynote demo. Two patents followed. NSX Intelligence became a reference pattern for how VMware approached ML-driven security going forward.
NSX Intelligence: VMworld 2019 LaunchWATCH ON YOUTUBE
79%
Approval rating
Post-keynote attendee survey · VMworld 2019
18%
Conversion rate
From demo booth to qualified opportunity
13
PoC sign-ups
Fortune-500 customers · first 30 days
65%+
NSX-T adoption
In the monitored migration window post-launch
We've been trying to get visibility like this for two years. This is the first time I trust what I'm seeing enough to act on it.
Network Admin, VMworld 2019
10 / ITERATIONFeedback loop
Every percentage became a design decision.
MY ROLEAnalyzed VMworld survey data; translated customer feedback into three specific design changes for the production release.
From the 79% approval and 18% conversion at VMworld, we dug into the why.
43%
Wanted to modify AI recs
Added a recommendation editor so admins could tweak before applying.
23%
Re-emphasized simulation
Made the simulation step more prominent in the workflow.
61.5%
Wanted fine-tuning controls
Added advanced controls + export for power users.
11 / REFLECTIONWhat it taught me
Designing with AI is designing with uncertainty.
Every lesson from this project traveled with me. The instinct to build trust in failure modes, not success modes. The instinct to make every AI action reversible. The instinct to treat a confidence number as a design element, not a label. Five years later, those same patterns are load-bearing in the Alkira AI Framework.
01
The abstraction paradox
Users said "give me more knobs" AND "this is too much detail" in the same session. Designing for expertise means holding both truths.
02
Building trust in AI is a progression
Skepticism, then transparency, then control, then validation, then adoption. You can't skip a step.
03
The ML foundation
This was 2019. ML-based recommendations. What I've been building at Alkira takes this further: generative AI, MCP pipelines, and organizational transformation.
