survive.

An AI-powered retention engine that cures schedule overloading and transition shock by giving students a predictive study planner and universities a proactive radar to stop dropouts.

Comment

Inspiration

We’ve all watched brilliant classmates lose their scholarships or drop out during their freshman year. The culprit is rarely a lack of intelligence; it is a systemic failure of executive function. We call it "Syllabus Shock." Universities hand 18-year-olds hundreds of pages of unstructured PDF syllabi on the first day of class and expect them to magically map out 16 weeks of colliding midterms, labs, and research papers. When those deadlines inevitably crash into each other around Week 8, learners hit a wall they simply cannot climb over.

We realized that while universities spend millions on reactive solutions—like tutoring centers and academic probation—no one was building a proactive system. We were inspired to build Survive to bridge the gap between a student's need for personalized executive support and a university's need to protect its students from preventable burnout.

We realized that while universities spend millions on reactive solutions—like tutoring centers and academic probation—no one was building a proactive system. We were inspired to build Survive to bridge the gap between a student's need for personalized executive support and a university's need to protect its students from preventable burnout.

What it does

Survive is a dual-sided retention engine built on public institutional data and LLMs.

For the Student, it acts as a frictionless executive function coach. By uploading their syllabi, our engine parses their entire semester and visualizes it on a predictive Workload Heatmap. If the system detects a severe deadline overlap, our AI Triage Advisor prescribes a policy-aware survival strategy, and our Schedule Optimizer dynamically flattens their flexible assignments to create breathing room.

For the University Administration, the app features a hidden "God-Mode" (accessed via Shift + A). The Campus Burnout Radar aggregates cohort data to show Deans exactly which courses are causing structural collisions at scale, calculating the exact tuition revenue at risk and allowing them to proactively alert faculty before students fail.

How we built it

We architected Survive to be highly resilient and deeply integrated with public data, specifically targeting the University of Michigan LSA Syllabus Archive. The core of our ingestion pipeline is a live Node.js Express microservice. Unstructured PDFs and DOCX files are routed to our backend, where an LLM heuristics engine (gpt-4o-mini) processes the documents entirely in-memory. To calculate the structural risk, we built a weighted heuristic algorithm. High-stakes assignments (like midterms and final projects) receive heavy threat weights, while standard assignments receive lighter weights. The engine aggregates these across a 16-week matrix to output the student's Survival Score, utilizing the following bounding function to penalize structural collisions:

$$S = \max(15, \min(85, 80 - 15c))$$

(Where 'S' is the final Survival Score out of 100, and 'c' is the number of Critical Collision Weeks detected in the timeline).

The frontend is a vanilla JavaScript single-page application built for zero-latency UI updates, featuring dynamic SVG rendering and a custom .ics string-builder that allows students to export their optimized schedules to Google Calendar without heavy OAuth flows.

Challenges we ran into

Extracting chronological data from highly localized, unstructured college syllabi is notoriously difficult. A grading table in a History syllabus looks entirely different from a bulleted reading list in a Computer Science syllabus.

To solve this, we engineered a three-tier fallback architecture:

The app first attempts to parse the document using our backend LLM pipeline to return structured JSON.

If the AI fails or the structure is malformed, the server seamlessly falls back to a custom Regex heuristic engine.

If the server is unreachable, the client-side JavaScript utilizes a Web Worker (pdf.js) to process the document locally in the user's browser.

This redundancy ensures that the application remains functional even under heavy load or API rate limits.

Accomplishments that we're proud of

The "Trojan Horse" Architecture: We successfully bridged the UI/UX gap between a consumer-friendly student study app and a highly valuable enterprise administrative dashboard.

Policy-Aware AI: Instead of generic advice, our AI Triage engine provides hyper-realistic coaching, such as warning students when they do not meet a university's specific 3-exam reschedule policy.

In-Memory Processing: We achieved 100% FERPA compliance by ensuring our Node.js pipeline processes all unstructured document data in-memory, without ever storing PII in a database.

What we learned

We learned the massive distinction between building a tool for the end-user (the student) versus building a product for the buyer (the university). From a technical perspective, we gained deep experience in document parsing pipelines, learning how to wrangle pdf-parse, mammoth, and LLM system prompts to reliably extract structured temporal data from chaotic real-world academic PDFs.

What's next for survive.

Our next major step is integrating Survive directly into University Learning Management Systems (like Canvas and Blackboard) via LTI (Learning Tools Interoperability). This will eliminate the need for manual syllabus uploads entirely, allowing the Campus Burnout Radar to run silently in the background from day one of the semester, protecting students before they even know they are at risk.

Built With

Share this project:

Updates