GuardAI – AI Risk Management Framework
Contributed to the redesign of GuardAI, a risk management platform aimed at helping organizations safeguard their generative AI services against misuse and align with evolving US/EU compliance standards.
Role
UI/UX Designer
Duration
3 Months
Tools
Figma
Company
Himalaya Quantitative Solutions
About
GuardAI is an AI risk management framework developed by Himalaya Quantitative Solutions. It protects generative AI systems from harmful or malicious usage, helping developers and businesses comply with evolving US and EU regulations.
As a volunteer UI/UX designer, I redesigned GuardAI’s public-facing interface to improve product clarity, usability, and visual credibility—directly supporting client acquisition and product demos.
Challenge
The original GuardAI interface lacked credibility and structure, making it difficult for users and potential business clients to understand its capabilities. This posed a barrier for the team’s goal: showcasing the platform’s ability to prevent misuse of GenAI (e.g., jailbreak prompts, harmful content) and help enterprise clients meet regulatory standards.
Contributions
Redesigned the GuardAI website to improve structure, clarity, and professionalism, with emphasis on compliance, use cases, and AI safety messaging.
Improved visual hierarchy and navigation to better communicate product value (e.g., Defense, Risk Metrics, Use Cases).
Collaborated closely with engineers to align on feasibility and implementation timelines.
Enhanced product demos used in business presentations, helping improve engagement with potential clients.
Homepage
Redesign: Establishing Trust from the Start
To help users and stakeholders understand the vision behind GuardAI, I designed an About page that:
Establishes the problem space: security vulnerabilities in AI
Explains how GuardAI protects against malicious misuse
Communicates the ethical responsibility of building secure models
About: Framing
the Why
To help users and stakeholders understand the vision behind GuardAI, I designed an About page that:
Establishes the problem space: security vulnerabilities in AI
Explains how GuardAI protects against malicious misuse
Communicates the ethical responsibility of building secure models
Case Study: Transparency in Action
We created a public-facing case study to share real use cases and evaluation outcomes. It allows users to:
View examples of prompt injections and data leaks
Understand how defenses perform in different scenarios
Build credibility and share research with others
Vulnerability Detection &
Defense
⚠️ Harmful Prompt Tab: Exposing Prompt Injection Risks
This tab allows users to test how easily models can be manipulated.
Input: Email, model, original prompt, # of attack variations
Defense dropdown: Choose mitigation method
Attack preview: See how attackers modified the original prompt
Real-time feedback: Did the model follow the attack prompt?
🔒 Privacy Tab: Preventing Data Leakage
Tests if the model can reveal sensitive information it's exposed to.
Inputs: Email, model, prompt
File reference: Simulates model access to private files
Output reveals if sensitive data was unintentionally surfaced
🟰 Bias Tab: Detecting Fairness and Equity Issues
This tab allows users to test model outputs across identity-based prompts.
Side-by-side comparisons with demographic variations
Output is scored for inconsistency or skew
Designed to flag ethical concerns in hiring, healthcare, etc.
