DeepSeek vs. Llama: The "Kill The Cloud" Challenge

DeepSeek vs. Llama: The Kill The Cloud Challenge Cloud AI dominates benchmarks, but real-world apps need privacy, speed, and offline reliability. This challenge is about imagining what AI looks like when it runs entirely on-device. Hosted by Singularity, and powered by RunAnywhere, this challenge explores the future of privacy-first, offline, on-device AI using modern models like DeepSeek and Llama. It’s an ideation-focused challenge. No heavy coding required — bring your ideas, system thinking, and vision for how AI should truly work for users. Join the WhatsApp group for updates and communication: https://chat.whatsapp.com/LaPkZOMpmN97A7CXcVJEtu Platform & Communication: This is an online hackathon. Submissions will be made on Unstop Final judging and interactions will take place on Discord Official announcements and updates will be shared via WhatsApp Participants must also join the Discord server, as the final judging will take place on Discord: https://discord.gg/N359FBbDVd Eligibility: Open to everyone Students, working professionals, indie developers, designers, founders, and self-taught builders are eligible No restriction on academic background, degree, or specialization Participants from all domains and disciplines are welcome No prior experience with AI or mobile development is mandatory Team Guidelines: Participation is allowed only in teams Team size must be between 2 and 4 members Inter-college teams are allowed Inter-specialization and cross-domain teams are allowed Team members do not need to belong to the same organization, institution, or location Challenge Format: This is a concept and architecture-based ideation challenge Teams are expected to propose a mobile app idea that works fully offline using on-device AI Submissions should clearly explain: The problem being solved Why on-device AI is essential for this use case How the app would function without cloud APIs No executable application or deployed product is required Problem Focus Areas: Privacy-First Applications: Use cases where sensitive user data should never leave the device (for example: health, finance, personal journals, identity, security) Offline-First Experiences: Applications designed for environments with poor or no internet access, such as remote locations, transit systems, or emergency scenarios Low-Latency Intelligence: AI-powered experiences that require instant responses by running models locally instead of relying on remote servers Submission Expectations: Submissions should demonstrate clear thinking, feasibility, and originality Focus on architecture, workflows, and model usage rather than surface-level ideas Marketing-heavy or vague descriptions should be avoided Each team must submit one consolidated entry Rules: Teams must adhere to the 2–4 member limit Each participant can be part of only one team All submissions must be original work created specifically for this challenge Plagiarism or reuse of existing projects without significant modification may lead to disqualification Decisions made by the organizers and judges will be final