Inspiration
College admissions officers look for resilient, high-potential students, but raw numbers like GPAs and AP course counts rarely tell the whole story. A student who takes 4 AP classes at a heavily funded school is vastly different from a student who takes 4 AP classes at a school that only offers 5, all while working 20 hours a week to support their family. We realized these numbers cannot truly be measured without environmental context. We built DeltaMetrics to quantify this context, leveling the playing field for under-resourced applicants by surfacing the unseen barriers they have overcome.
What it does
DeltaMetrics is an AI-powered assessment dashboard that generates a detailed account of an applicant's background and the context of their schooling. When an admissions officer inputs a student's high school, ZIP code, APs taken, and their non-academic "survival hours", (such as required jobs, sibling care, family translation, or elder support), the platform dynamically researches the school's resources in real time. It visualizes critical data, such as the student-to-counselor ratio, community median income, and the number of AP/Honors courses taken versus courses actually offered. Finally, it leverages AI to synthesize this quantitative data into a highly objective, 150-word professional summary of the applicant's background environment, ensuring admissions teams can make equitable and context-aware decisions. A student from a more affluent area wouldn't have to waste words on their personal statements discussing the time they spent caring for their family; instead, they would have more words to discuss their academics and interests in relation to the school they're applying to. Our website helps overcome this massive barrier.
How we built it
We architected a full-stack web application designed for speed and live data retrieval. Frontend: Built with React and styled with Tailwind CSS, deployed on Vercel. It features dynamic charts and a responsive, data-dense UI. Backend: A robust Python FastAPI server deployed on Render, utilizing strictly typed Pydantic models for data validation and CORS middleware for secure cross-origin communication. AI Integration: Instead of relying on static, hardcoded databases, we integrated the Google Gemini API (model gemini-3.8-flash) utilizing Google Search Grounding. This allows the backend to autonomously search the web for accurate, real-world statistics about any high school in the country, guaranteeing the data is fact-checked and hallucination-free before synthesizing the final report.
Challenges we ran into
Building a complex, multi-service deployment during a hackathon tested our endurance. We faced significant hurdles with our cloud infrastructure, including battling CORS preflight errors between Vercel and Render. Our biggest challenge was migrating from a local, mock database to a live AI search agent. We had to relentlessly debug API rejections (502 and 402 errors), manage strict payload schema mismatches (snake_case vs camelCase), and carefully engineer our Gemini prompts. We had to ensure the model utilized Google Search Grounding to find accurate statistics while simultaneously forcing it to output strictly formatted JSON strings so our React frontend could properly map the data to our charts.
Accomplishments that we're proud of
We are incredibly proud of building a fully deployed, completely functional product from end to end. We didn't just build a mock UI; we engineered a live pipeline that successfully bridges a React frontend, a Python FastAPI backend, and an advanced AI agent. We successfully implemented live web searching to completely eliminate any AI hallucinations, as it forced the model to act as a rigorous data researcher rather than simply generating random statistics.
What we learned
We gained hands-on experience in full-stack debugging and cloud orchestration. We learned how to read through dense server logs to pinpoint API configuration errors, how to properly handle asynchronous state management in React, and how to combine LLM reasoning with live search tools. Most importantly, we learned that building equitable technology requires rigorous data validation, as you cannot rely on AI to guess the facts when a student's future is on the line.
What's next for DeltaMetrics
Moving forward, we want to expand DeltaMetrics beyond Google Search by directly integrating the National Center for Education Statistics (NCES) government database API for flawless, standardized baseline data. We also plan to build batch-processing capabilities, allowing university admissions offices to upload thousands of applicants via CSV to generate context reports at scale, seamlessly integrating with platforms like the Common App.
Built With
- fastapi
- gemini
- javascript
- python
- react
- vercel
