Transforming Campus Innovation with an AI Platform

Building an Adaptive, Accessible AI Platform at Cornell University

Overcoming Barriers to AI Adoption

• Cornell University faced common challenges around AI adoption, including:
  • Lack of a centralized, safe place for experimentation
  • Difficulty integrating AI into existing workflows and processes
  • Concerns around cost, data protection, and privacy

Empowering Citizen Developers

• Instead of a centralized AI team, Cornell took a different approach:
  • Provided tools and platform to enable anyone on campus to build AI solutions
  • Believed AI should be accessible like email or Slack, not a specialized project
  • Focused on reducing barriers to entry and making the platform easy to use

Key Components of the AI Platform

1. Light LLM: AI gateway that provides access to foundation models through AWS Bedrock
2. N8N Agent Studio: Low-code workflow automation platform for building AI-powered applications
3. Libra Chat: Conversational interface built on top of the AI gateway

Technical Architecture and Deployment

• Leveraged AWS services like CloudFormation, CodeBuild, CodePipeline, and ECS Fargate
• Automated infrastructure deployments and CI/CD workflows
• Enabled rapid updates to take advantage of new AI model releases

Use Case 1: Adaptive Student Values Exploration

• Developed a chatbot to help incoming engineering students explore their values and goals
• Allowed students to engage with the AI assistant before meeting with a human coach
• Resulted in over 40,000 student interactions with zero support requests

Use Case 2: Socratic Chat for Active Learning

• Built an AI-powered Socratic chat application integrated into the learning management system
• Allowed professors to gauge student understanding in large classes beyond just facial cues
• Provided students with a guided, structured learning experience based on Bloom's Taxonomy
• Enabled instructors to analyze conversation data and identify areas needing more attention

Use Case 3: Automated Expense Reimbursement

• Addressed a backlog of 10,000 expense reimbursement requests per semester
• Used AI-powered automation to extract and analyze receipt data, apply policies, and provide recommendations
• Saved over 30 minutes per reimbursement request, allowing the team to focus on more strategic work

Fostering a Culture of Innovation

• Engaged the campus community through open office hours and student projects
• Empowered "citizen developers" to experiment and build solutions using the AI platform
• Collaborated with other universities to share best practices and drive innovation in higher education

Key Takeaways

• Providing a centralized, accessible AI platform enabled rapid innovation and adoption across Cornell's campus
• Focusing on reducing barriers and empowering end-users (rather than a centralized AI team) was a key success factor
• Leveraging AWS services like Bedrock, Fargate, and CI/CD tools allowed for efficient, scalable deployment of AI applications
• Specific use cases demonstrated the platform's impact on student engagement, active learning, and operational efficiency
• Fostering a culture of innovation and collaboration was crucial for driving widespread AI adoption