Catapult | AI x Hackathon | Purdue 2025
In today’s fast-paced work environments, it’s easy to forget what was said, decided, or discussed during meetings, lectures, and personal conversations. We take notes, but they're often incomplete or difficult to understand. We record audio, but rarely go back to listen. What if you had an intelligent memory system that could listen, understand, and remember - like your very own personal assistant?
We built Neuron to capture that idea. It not only transcribes your meetings, but also generates instant meeting summaries and intelligently tags key ideas for fast retrieval, all without needing the cloud and kept completely hands free. Neuron knows your schedule, listens in on those events, generates high quality markdown notes, catches the details you miss. We all would love our own personal assistant, and Neuron ensures your thoughts stay private, accessible, and actionable so that you can focus on what really matters.
Neuron is a mobile-first AI app that helps users capture, summarize, and search their daily thoughts and meetings. Some of the key features of this app:
Records audio from meetings and converts it to text in real time Summarizes transcriptions into concise, readable meeting notes Generates vector embeddings to represent the each note semantically Assigns context-aware tags to the note for easy search and filtering Works completely offline, ensuring privacy and confidentiality, as well as low latency Calendar integration and event creation capabilities with your phone's Google Calendar
We designed various algorithms and implemented models within this app:
Flutter was used to develop the clean and responsive mobile UI Whisper (from OpenAI) was leveraged for offline yet robust audio transcription Fine-tuned Llama3.2:1b for on-device summarizations of transcriptions Vector embeddings are generated using sentence-transformers (all-MiniLm-L6-v2), and stored locally in a vector database
📜 Offline Transcription using Whisper
✂️ Summarization for digestible meeting notes
🧩 Auto-tagging using contextual NLP
🔍 Semantic Search via vector embeddings
📴 Offline-first Architecture for full privacy
Balancing performance vs. accuracy in summary and embedding generation on mobile hardware Ensuring the embedding and tagging pipeline worked consistently across a wide range of topics Accomplishments that we're proud of
Successfully running end-to-end vector search and summarization offline on a mobile device Implementing a custom tagging system that adapts to any domain — no need for predefined labels Calendar system that seamlessly integrates with your Google Calendar, allowing users to create, edit, and delete right from the app
Running even small LLMs on-device requires careful optimization and consideration of hardware requirements Embeddings offer more than search — they're the key to building memory graphs and context linking What's next for Neuron
Stay tuned for the next updates of Neuron. We plan to continue enhancing it so that it can improve your life. Here are some sneak peaks of our upcoming features:
Back up data securely on cloud as protection against loss of local storage Add a focus directed graph to visually display various links between notes Automatically schedule recordings based on your calendar, so you never miss a meeting Attach images, screenshots, and sketches to your memory timeline for richer context
https://devpost.com/software/neuron-k8ij9s?ref_content=user-portfolio&ref_feature=in_progress