Pilot GenAI

NYU is building its generative AI platform, Pilot. I joined in an early stage as an UI Developer.

Collaborated with 20+ departments across the campus including CAS, Steinhardt, Provost, Wagner and more, we developed multiple use cases and features to help NYU community to learn, work and research.

01. Context and Challenge: AI in Education, Trend? Future.

The Challenges Facing NYU (and Higher Ed)

1. Fragmentation Across Use Cases & Departments
   AI initiatives in universities often live in silos—e.g. separate pilots for tutoring, grading, or admin functions—leading to duplicated effort and inconsistent experiences.

2. High Stakes for Efficiency & Quality
   With AI saving teachers nearly half of their planning time and boosting engagement nearly 50%, systems that don’t integrate these gains universally fall short—especially in a large, complex institution like NYU.

3. Inconsistent Adoption & Ethical Guidance
   Around 80% of students use AI independently—but only ~20% of faculty feel prepared to implement it effectively, and concerns about fairness, bias, and integrity persist.

Answer: a Access-Unified, Multi-Stakeholder, Trustworthy and scalable AI Platform

02. Objectives & Learning Metrics: Instructor-Student Co-Feedback

What we set out to achieve

Pilot GenAI aims to unify fragmented AI efforts across NYU—spanning learning, teaching, administration, and research—into a single, trustworthy platform that scales across schools and departments.

Design Objectives:

• Clarity & trust: Make AI interactions legible, guided, and safe for students and faculty.

• Speed to value: Reduce time from question → useful answer across tutoring and workflow tools.

• Scalable patterns: One design system that supports 20+ use cases without bespoke UIs each time.

• Seamless adoption: Lower friction for departments to pilot, configure, and roll out features.

Learning Metrics:

Since Pilot GenAI is in its pilot phase, we don’t yet track DAU/MAU as primary success metrics. Instead, our focus is on gathering qualitative feedback and early engagement signals from teachers and students across CAS, Wagner, Provost, and Steinhardt. I designed and led multiple feedback loops (testing sessions, surveys, faculty workshops) to surface recurring themes, which now drive iteration. Our goal is to translate these insights into measurable adoption and engagement KPIs as the platform scales.

03. Research & Discovery: Acknowledging AI in College

Starting Point: What does each role in NYU community need from AI?

Before designing interfaces, I worked with multiple NYU departments (CAS, Wagner, Steinhardt, Provost, and more) to surface real problems students, teachers, and administrators face daily.

• Students wanted AI that went beyond just answers—tools that explain and teach

• Faculty needed relief from repetitive workflows (grading, recruiting, research templates, files).

• Administrators & researchers sought structured, compliant support for grants, costs, and publications.

Because Pilot GenAI is still in testing, feedback became the most valuable currency.

• Ran feedback workshops and demo sessions with teachers and students.

• Collected pain points and feature requests.

• Synthesized feedback into affinity maps and design priorities.

Insights That Shaped Design of Pilot GenAI:

• Explainability > Accuracy alone: Students valued step-by-step clarity even more than the final answer.

• Consistency across use cases: Teachers wanted one platform to handle both tutoring and admin tasks, not a patchwork of tools.

• Trust & transparency: Faculty stressed the importance of showing how AI derived answers, especially for grading or research contexts, and getting into the students' perspective towards AI.

04. Design & Branding: AI Chat. Group Chat.

Design Goals

From research with students and faculty, three guiding principles shaped the design of Pilot GenAI:

• Clarity & trust → Interfaces must make AI outputs explainable and usable in academic contexts.

• Consistency across use cases → One design system that works for tutoring, administration, and research.

• NYU identity → Visually aligned with NYU’s theme while introducing AI-native interaction patterns.

UI System & Structure

One of the core contributions was structuring the platform’s UI to reflect how NYU itself is organized:

• Scalable UI for Campus: Pilot GenAI is accessible to all NYU community, and is based on generative AI, staring with chats.

  

• Group-Based Access: The platform mirrors classes, departments, and roles. Instructors manage student groups, assign access, and monitor usage.

  

• Teacher / Lead as Administrator: The administrator will have the authority of the group, depending on their presets managed by super administrator (NYU IT). They can add/drop, share model and files in the group.

  

• Shared Conversations: Student–AI dialogues can be part of assignments—reviewed, shared, or assessed by instructors.

  

  
  

• Instructor Oversight: Faculty can see conversations, provide guidance, and co-develop AI models with our team to fit their curriculum.

  

• Platform Familiarity: Styled like a modern AI platform (similar to ChatGPT) but grounded in NYU’s academic brand.

  

• Framework Quick View:

  

Visual Identity

• Logo: Designed a spiraling flower motif built from NYU’s iconic torch elements. It symbolizes growth, collaboration, and the expanding reach of AI in education.

  

• Color & Style: Applied NYU’s official palette to maintain brand consistency, balanced with modern gradients and soft edges to signal approachability.

• Motion Design: Created subtle animations to bring conversations and transitions to life without distracting from tasks.

  

05. Prototyping & Testing - Use Case: Math Ally

Why Math Ally?

Math is one of the toughest areas for students to learn independently — they often either get just the answer (no learning) or get lost in abstract explanations. Math Ally was our pilot to test how AI could teach through dialogue while still fitting into NYU’s class structures, and planned for future teaching scenarios.

What We Built

• Guided Questioning → Instead of giving direct answers, Math Ally breaks down problems into key sub-questions that lead students to solve step by step.

• Instructor Dashboard → All conversation histories are logged and summarized into quantitative insights (e.g., time spent per question, number of hints requested).

• Class-Specific Models → Each homework set (≈8 per semester, plus midterm and final) gets a model co-trained with teachers and our team — improving each term.

• Python Execution → Every solution can run in Python, ensuring precision and verifiability.

• Assignments by Conversation → Conversations themselves become a core part of homework submissions, letting instructors assess process as well as outcome.

Testing & Feedback Loops

To validate these designs, I worked directly with math instructors and students across CAS:

• Pilot sessions: ran with sample homework problems, logging how students engaged with sub-questions.

• Instructor workshops: collected feedback on what data they found useful vs. overwhelming in the dashboard.

• Iteration cycles: adjusted UI to emphasize progress indicators and summary metrics (e.g., “average time per step”) instead of raw logs only.

My Impact in This Phase

• Researched and designed the full UI for both student conversations and instructor dashboards.

• Implemented frontend prototypes for conversation flow and history visualization.

• Designed and ran testing scenarios with instructors + students, collecting feedback and iterating.

• Validated Python execution integration as part of testing for accuracy.

• Synthesized teacher feedback into requirements for the next term’s homework/exam integration.

06. Impact & Early Outcomes: AI, Scaling Up

Adoption Across NYU

• Pilots already running in multiple departments:

  • Math Ally and Linear Algebra Tutor will be applied in 2025 Fall semester.

  • JD Writer and Allowable Costs are applied and testing now.

• Other 20+ use cases designed, ranging from math tutoring to HR support, research workflows, and language learning.

• Working with more departments to develop more use cases.

Roadmap: Teaching, Working and Researching with AI

• 2025 Fall Semester (Teaching)

  • Launch Math Ally and Linear Algebra Tutor (LAT) in math classes.

  • Integrate conversation-as-assignment into homework, midterms, and finals.

  • Expand reading prompt evaluation and language-learning chatbots to support non-English speakers.

• Administrative Workflows (Working)

  • Scale up JD Writer to streamline job description creation across departments.

  • Deploy Allowable Costs Manager for grant and budget validation.

  • Roll out Career Pathways to provide resume feedback and job-tailored optimization.

• Research Support (Researching)

  • Implement Research Template Generation to automate recurring paperwork.

  • Extend Research Publication Search & Classification for linked knowledge discovery.

  • Develop AI assistants to support metadata collection for e-books and multilingual research.

• Long-Term Scaling

  • Co-develop models with instructors each semester, improving accuracy and curriculum alignment.

  • Create a unified AI ecosystem where teaching, administration, and research tools are accessible through one consistent platform.

  • Establish feedback-driven iterations with students, faculty, and administrators to refine models and UX every term.