Collaborating on team project “Computational Modeling of Parasocial Attachments and Social Simulation”.

• Selected as one of 300+ mentees from 1600 applicants in the largest AI alignment research program in the world.

• Conducted extensive literature review on LLMs’ “theory of mind”, critical for understanding AI deception capabilities. • Co-designed novel game-based paradigm to measure new aspects of LLM theory of mind.

• Co-developed Python code to test dozens of different LLMs through OpenRouter API.

• Executed 4680 trials across 6 frontier models (Sonnet 4.5, GPT-5, Gemini 3, Llama 405B) and 39 scenario types.

• Analyzed data and discovered key results with implications for real-world safety scenarios.

Self-directed intensive study program on AI safety, with emphasis on technical research applications.

Independent research project:

• Designed and executed experiment testing hypothesis that LLMs learn misaligned behaviors from sci-fi "rogue AI" tropes in training data

• Ran 400 trials in a 2x2 experimental design with ChatGPT5 and Claude Sonnet 4 using OpenAI and Anthropic APIs.

• Implemented automated LLM scoring system with 93% inter-rater reliability.

• Published research on LessWrong: “Misalignment and Roleplaying: Are Misaligned LLMs Acting Out Sci-Fi Stories?” (https://www.lesswrong.com/posts/LH9SoGvgSwqGtcFwk/misalignment-and-roleplaying-are-misaligned-llms-acting-out)

Leadership:

• Co-organizer, then Lead Organizer, for the BlueDot AI Safety Evals Reading Group (https://evalsreadinggroup.com), with ~50 attendees each week. Presented, led discussions, arranged author presentations, planned strategy.

Coursework and reading:

• Completed foundational AI coursework, including "AI for Everyone" (Coursera), "AI Essentials for User Experience Designers" (LinkedIn Learning), and “The Future of AI” (BlueDot.org)

• Completed specialized courses ”AI alignment” (BlueDot.org). “Practical Deep Learning for Coders” (fast.ai), “Machine Learning Crash Course” (Google), and “UI/UX Design for AI Products” (Stanford University).

• Engaged in extensive self-study of AI alignment literature, including academic papers and industry research.

• Earned certification in Python from Learnx.org.

• Fine-tuned and deployed model for predicting LLM response safety (https://huggingface.co/spaces/keavney/LLMSafetyClassifier).

Led design at Clumio, a software startup doing enterprise backup for AWS.

• Oversaw the delivery of high-quality UX designs for dozens of projects, including new flagship products used by Atlassian and Duolingo to back up petabytes of data and tens of billions of S3 objects.

• Contributed to Clumio’s outstanding net promoter score of 86. Ease of use one of our top selling points and was routinely cited as a competitive advantage by our sales force.

• Annual revenue grew from <$1M to $25M during the time I was leading the UX team.

Built a world-class UX design team:

• Hired 7 designers with skills strategic for the business in a tight job market. Team retention was 100%.

• Established leveling and career tracks for UX job roles. Every designer understood the expectations of their role and had SMART performance goals.

• Built team spirit by coaching and mentoring employees. My team had a CultureAmp employee engagement score of 100 (overall company score was 64).

Established processes and best practices that led the team to excel:

• Increased productivity by streamlining UX tool use from three different tools to one and creating a rich Figma design system library of complex branded components and page templates.

• Ensured consistency by creating a complete set of design system guidelines with foundational elements like color and typography, usage rules and code samples for each component, voice and tone standards, and UX patterns.

• Established accountability by setting up a formal UX development process that integrated into Clumio’s software development lifecycle.

• Improved quality by Integrating user research into all aspects of the design process, from analytics to understand product use, foundational user interviews to discover user needs, and usability testing to measure success.

Designed consumer-facing UI for travel application