An AI agent that turns a wall of unresolved security alerts into plain-language context, ranked priorities, and one-click remediation.
An AI agent that turns a wall of unresolved security alerts into plain-language context, ranked priorities, and one-click remediation.
An AI agent that turns a wall of unresolved security alerts into plain-language context, ranked priorities, and one-click remediation.
In labs
As Product Designer, worked on WebUI app.
As Product Designer, worked on WebUI app.
As Product Designer, worked on WebUI app.
Tools used:
Tools used:
3rd place in Egnyte Hackathon 2025, Best Use of AI
3rd place in Egnyte Hackathon 2025, Best Use of AI
3rd place in Egnyte Hackathon 2025, Best Use of AI
50% of all security issues detected in the last 12 months were never acted upon. Not because admins don't care but because the system gives them noise, not signal.
50% of all security issues detected in the last 12 months were never acted upon. Not because admins don't care but because the system gives them noise, not signal.
50% of all security issues detected in the last 12 months were never acted upon. Not because admins don't care but because the system gives them noise, not signal.
Research & Exploration
Research & Exploration
Research & Exploration
We started with ethnographic research and behavioral analysis using Mixpanel and Pendo to understand why existing security workflows had low adoption despite high issue volumes. The goal wasn’t just to surface more alerts, but to reduce noise and help users take meaningful action faster.
The AI agent experience was shaped through close collaboration with developers to define how recommendations, remediation flows, and issue clustering should behave within the product. Concepts and interaction models were iteratively refined through usability testing with customers before broader rollout.
If you'd like to discuss the product thinking, research, or design decisions behind this project, let’s connect.
We started with ethnographic research and behavioral analysis using Mixpanel and Pendo to understand why existing security workflows had low adoption despite high issue volumes. The goal wasn’t just to surface more alerts, but to reduce noise and help users take meaningful action faster.
The AI agent experience was shaped through close collaboration with developers to define how recommendations, remediation flows, and issue clustering should behave within the product. Concepts and interaction models were iteratively refined through usability testing with customers before broader rollout.
If you'd like to discuss the product thinking, research, or design decisions behind this project, let’s connect.
We started with ethnographic research and behavioral analysis using Mixpanel and Pendo to understand why existing security workflows had low adoption despite high issue volumes. The goal wasn’t just to surface more alerts, but to reduce noise and help users take meaningful action faster.
The AI agent experience was shaped through close collaboration with developers to define how recommendations, remediation flows, and issue clustering should behave within the product. Concepts and interaction models were iteratively refined through usability testing with customers before broader rollout.
If you'd like to discuss the product thinking, research, or design decisions behind this project, let’s connect.
Key Product Decisions
Key Product Decisions
Key Product Decisions
1 | AI Analysis & Recommendation providing a plain-language context on demand.
1 | AI Analysis & Recommendation providing a plain-language context on demand.
1 | AI Analysis & Recommendation providing a plain-language context on demand.
Problem: Users saw raw issues details with no guidance on what it meant or what to do. Less experienced teams struggled to interpret and act on the issues confidently.
Problem: Users saw raw issues details with no guidance on what it meant or what to do. Less experienced teams struggled to interpret and act on the issues confidently.
Decision: An on-demand AI button generates a plain-language narrative: what triggered the issue, who's involved, and a specific recommendation to remediate, dismiss, or delegate.
Decision: An on-demand AI button generates a plain-language narrative: what triggered the issue, who's involved, and a specific recommendation to remediate, dismiss, or delegate.
Why: On-demand keeps AI advisory. Users stay in control and build trust gradually rather than feeling overridden.
Why: On-demand keeps AI advisory. Users stay in control and build trust gradually rather than feeling overridden.
2 | AI Configuration Agent enabling a natural language policy creation
2 | AI Configuration Agent enabling a natural language policy creation
2 | AI Configuration Agent enabling a natural language policy creation
Problem: Complex forms and intricate policy logic deter most customers from activating preventive controls at all.
Problem: Complex forms and intricate policy logic deter most customers from activating preventive controls at all.
Decision: Users describe their intent in natural language. The agent generates a ready-to-review policy, inactive by default, requiring explicit user approval before it takes effect.
Decision: Users describe their intent in natural language. The agent generates a ready-to-review policy, inactive by default, requiring explicit user approval before it takes effect.
Why: The review step is non-negotiable, even if it adds friction. Incorrect policies could unintentionally block legitimate access.
Why: The review step is non-negotiable, even if it adds friction. Incorrect policies could unintentionally block legitimate access.
3 | AI to detect patterns, not just per-issue assistance.
3 | AI to detect patterns, not just per-issue assistance.
3 | AI to detect patterns, not just per-issue assistance.
Problem: Users miss patterns across thousands of open issues (shared root causes, bulk-closeable low-risk items, or correlated threats that individually look minor)
Problem: Users miss patterns across thousands of open issues (shared root causes, bulk-closeable low-risk items, or correlated threats that individually look minor)
Decision: The agent proactively surface issue clusters, recommending bulk actions.
Decision: The agent proactively surface issue clusters, recommending bulk actions.
Why: Issue fatigue is a problem. Solving it per-issue doesn't entirely help in resolving 70% high severity 'Unusual access' issues (for eg.).
Why: Issue fatigue is a problem. Solving it per-issue doesn't entirely help in resolving 70% high severity 'Unusual access' issues (for eg.).
The hardest thing wasn't designing the AI. It was deciding how much of the decision to give it, and how much to keep with the human.
The hardest thing wasn't designing the AI. It was deciding how much of the decision to give it, and how much to keep with the human.
The hardest thing wasn't designing the AI. It was deciding how much of the decision to give it, and how much to keep with the human.
© 2026 All Rights Reserved | Parmi Mehta
© 2026 All Rights Reserved | Parmi Mehta