Built at the first-ever Penn Generative AI Hackathon, SitDowns is a Chrome extension that replaces exhausting team standups with daily summaries of you and your team's web-browsing activites. When enabled, the Plasmo extension scrapes the websites you visit and queries a Mixtral model served on Baseten, summarizing each page and synthesizing your daily rundown to upload to Supabase. User summaries are then sorted through Nomic embedding vector similarity search to display the most relevant teammate activites in list and graph forms.

Capybara.AI is the world’s most advanced AI college consultant. Say goodbye to GPT hallucinations and private college counseling gatekeeping, and generate better long-term recommendations than human counselors at 1% of the cost. Powered by a custom GPT-4 agent pipeline and Pinecone embeddings, the chat model analyzes specific needs to generate individualized tips and strategies. With access to vector memory, web access, and document upload capabilities, the chat agent remembers user interactions stored in Supabase and chooses the best tool for every interaction.

Focal is a comprehensive productivity solution built with Flutter that is available on Android and iOS. Start your day by adding tasks synced across all devices through Firebase. Instead of checking a box to complete a task, users will enter Focus Mode to commit to their work and track time spent on each task with either a traditional stopwatch or a smart Pomodoro timer. In Focus Mode, if the user gets distracted and uses another app, Focal's native mobile API integrations track what you are doing and notifies you. Once tasks have been completed, statistics provide daily reports with data from each completed task and monthly productivity trends.

Built at Penn Apps XXIV, RajivAI is a multi-agent model designed to help professors and students generate exams. Simply upload course materials such as lecture notes, specify topics, length, and types of questions, and the model will generate a full-length exam akin to human-written materials. A GPT-4 professor agent chunks the source material and delegates GPT-3.5 TA agents to summarize each part. With the whole course summarized, the professor then delegates TAs to collaborate to recursively cross-reference questions and answers to create a valid exam.