What I Built
I built Closure, an AI-assisted personal workspace designed to help manage "open loops."
I built this for a close friend and fellow university student who constantly juggles a dozen different things—hackathons, academic research, LeetCode grinding, and side projects. Standard to-do lists never worked for them because their work doesn't happen in a linear checklist; it happens in bursts across different platforms. This led to severe task paralysis and the anxiety of "open loops" taking up mental RAM.
Closure solves this by passively tracking what they are actually doing. Instead of forcing them to manually check boxes, Closure uses a custom browser extension and GitHub integration to observe their activity, uses AI to figure out which project that activity belongs to, and automatically organizes their workspace. It even ingests their course syllabi to find deadlines for them.
Code
Aaryaman-147
/
closure
An AI-assisted workspace that passively tracks background activity to organize projects, extract deadlines, and close "open loops."
Closure
Built for the Hacktoberfest Weekend Challenge: Build for a Friend, Closure is an AI-assisted personal workspace that passively tracks background activity to organize projects, extract deadlines, and close "open loops."
Traditional to-do lists fail for chronically busy students and developers because they require manual, rigid upkeep. Closure solves this by observing actual work (GitHub commits, LeetCode practice, YouTube research) and autonomously categorizing it into an actionable desk layout.
✨ Features
- The Relevance Engine: A background AI pipeline that maps raw activity signals (browser tabs, commits) to your active projects, or pushes brand new discoveries straight to your Inbox.
- Brain Dump Extraction: Type a messy paragraph of thoughts, and Gemma 4 structures them into categorized projects, recurring goals, and tasks.
- Passive Context Observer: Includes a custom Chrome Extension that quietly collects metadata from allowed domains (e.g., GitHub, LeetCode) without leaving the browser.
- Smart Deadline Extraction: Upload course syllabi, hackathon…
How I Built It
Closure is a full-stack Next.js application backed by MongoDB (Mongoose), heavily powered by Gemma 4 31B-it via the Google AI SDK.
Instead of just using AI for a chatbot, I engineered Gemma 4 to act as the core logic controller across several distinct backend pipelines:
- The Relevance Engine: A Chrome Extension passively sends metadata from allowed domains (LeetCode, GitHub, YouTube) to the backend. Gemma 4 evaluates this raw activity, decides if it relates to an existing project on the user's Desk, and creates an "Evidence Trail." If it spots a brand new project, it automatically drops it into the user's Inbox.
- Extraction Layer: Users can just "brain dump" a messy paragraph of thoughts, and Gemma structures it into categorized projects, recurring goals, and tasks.
-
Deadline Extraction: Using
pdf-parse, users can upload course syllabi or hackathon guidelines. The server extracts the text, passes it to Gemma 4 to hunt for critical dates, and automatically attaches deadlines to the workspace UI, bumping the project's attention level to "High." - Explainability: I built an Evidence UI that shows exactly why the AI flagged a specific task, citing the background signals it observed.
Why Does Open Innovation Matter?
Open innovation was critical for Closure. By using an open-weights model like Gemma 4, developers have the freedom to deeply integrate AI into background system architecture (like our Relevance Engine) rather than being boxed into consumer-facing chat wrappers.
Furthermore, the entire project relies on the open-source ecosystem—from pdf-parse for document ingestion to Next.js for the full-stack architecture. Open tools allow us to build highly personalized, privacy-respecting tools that adapt to how our brains actually work, rather than forcing us to adapt to rigid, closed-source productivity software.
Prize Categories
- Best Use of Gemma
Top comments (0)