GTA (Go To App): A KNN and Cosine Similarity Framework for Campus-Exclusive Skill-Based Peer Discovery and Real-Time Collaboration

University campuses contain diverse student skills, yet finding compatible teammates for projects, hackathons, and peer learning often depends on informal social networks. This paper presents GTA (Go To App), a campus-exclusive platform for skill-based peer discovery and real-time collaboration. Student proficiency is represented by weighted skill vectors, with weights of 0.3, 0.6, and 1.0 for beginner, intermediate, and advanced levels. A K-Nearest Neighbours engine uses cosine similarity to rank profiles and return the top 20 matches through a PostgreSQL Remote Procedure Call. Mandatory skill onboarding provides a non-empty profile for each new user. The platform combines institutional email authentication, mutual-acceptance messaging through Socket.io, campus communities, and events. A four-layer verification pipeline incorporates self-rating, timed quizzes, post-collaboration peer endorsement, and GitHub repository linking. The prototype uses React, Supabase, and AWS services. Evaluation involved 50 test student profiles and a database of 500 seeded profiles. All 50 profiles received a non-empty match list on their first request. Across 100 requests, mean search latency was 87 ms with a standard deviation of 23 ms. Deployment at 15,000-student scale and an annual infrastructure cost of approximately INR 20,750 remain projections. The framework integrates profile-based recommendations and communication within an authenticated campus environment.