-
-
Scholify
Inspiration
The idea for Scholify began after seeing several of my seniors and older friends get accepted into their dream universities but struggle to attend because of financial barriers. Many had worked for years to earn those acceptances, yet even after FAFSA and university scholarships, the cost was still far beyond reach. Some had to choose local in-state options instead, not because they lacked ambition, but because the right financial support never reached them in time. I noticed that many students either did not know where to look for private scholarships or found out about them only after deadlines had passed. That disparity was not something I could ignore. I realized that the problem was not a lack of opportunities, but a lack of accessibility and organization. I experienced this frustration myself while searching for scholarships. I found scattered databases, broken links, and confusing eligibility rules. The information existed, but it was disorganized and buried within outdated websites. As I dug deeper, I discovered hundreds of smaller private scholarships that supported different passions and backgrounds, yet most of them only reached a narrow group of students, typically those already connected to higher-income or well-networked circles. Many opportunities went unclaimed year after year simply because they never reached the students who needed them most. At the same time, I had been developing my skills in artificial intelligence and data engineering. I saw a clear way to bring those two worlds together by using AI not just as a technical tool, but as a way to make opportunities more equitable. I wanted to build something that could analyze every detail of a student’s profile and accurately match them to scholarships that truly fit their story. Scholify was born out of that goal. I decided to build my own structured database from the ground up, using an LLM-based multi-agent framework to curate the data into a structured schema and organize every piece of information so that it could be used intelligently by the algorithm. I designed the system to think like a mentor rather than a simple search engine, guiding students toward credible opportunities that might otherwise have remained undiscovered. My efforts were driven by a clear goal: to build a platform that brings clarity, fairness, and accessibility to the scholarship process for students from every background, helping them secure the funding they deserve and pursue their dream education without unnecessary financial barriers.
What it does
Scholify is a mobile app that helps students find scholarships they truly qualify for. It was built to solve a problem that affects millions of students every year: the process of searching for private scholarships is scattered, outdated, and often discouraging. Many great opportunities go unnoticed not because students lack ability, but because the right information is too obscure to reach. Scholify changes that by using artificial intelligence to turn scholarship discovery into a fast, accurate, and deeply personalized process. When a student completes the comprehensive onboarding process, Scholify stores a structured profile that captures their academics, extracurriculars, volunteer work, and personal background. This profile becomes the foundation for Scholify’s matching algorithm. Behind the scenes, the app analyzes the user’s data and compares it against a curated database of more than ten thousand verified scholarships worth over $5 million in total aid. Within seconds, it generates a clean, personalized feed showing the most relevant opportunities for that student. Each listing clearly displays the award amount, eligibility rules, deadlines, and sponsor details, allowing users to confidently choose from the best possible options. Scholify’s interface is designed to feel modern and intuitive. Students can scroll through their feed, bookmark scholarships they like, or apply directly from within the app. As they update their profile or save scholarships, the system instantly adjusts the feed to reflect their evolving interests and qualifications. Beyond discovery, Scholify includes an integrated AI assistant that helps students craft stronger applications. By reading both the scholarship prompt and the student’s profile, the assistant generates tailored outlines and ideas that highlight the student’s strengths and experiences. This feature is meant to help users overcome writer’s block and focus on what makes their story unique. Technically, the app is built as a full-stack solution using React Native for cross-platform access and a Python backend powered by FastAPI. A PostgreSQL database stores structured data, while a Pinecone database enables rapid similarity searches through the vectorized profiles and scholarship texts. The AI components are built with PyTorch and Hugging Face models, and the entire system runs securely on Google Cloud Platform. Scholify brings together design, data, and AI to make financial aid access more intelligent and equitable. It turns what used to be a stressful, uncertain process into one that is clear, personalized, and within reach for every student.
How we built it
We built Scholify as a full-stack, AI-powered mobile application.
Frontend:
React Native (Expo) for a cross-platform iOS and Android app
Modern component-based UI architecture
Backend:
Python with FastAPI for scalable, asynchronous API endpoints
Databases:
PostgreSQL for structured user profiles and scholarship metadata
Pinecone vector database for high-speed semantic similarity search
Cloud & Infrastructure:
Deployed on Google Cloud Platform
Secure authentication and data handling
AI/ML Stack:
PyTorch + Hugging Face
Sentence-BERT (SBERT) for semantic embeddings
Fine-tuned FLAN-T5 model
Retrieval-Augmented Generation (RAG) architecture for AI-powered application assistance
The system integrates structured data processing, vector similarity search, hybrid ranking logic, and LLM-based generation into a scalable production-ready architecture.
Challenges we ran into
Building Scholify was both the most rewarding and the most technically challenging project I have worked on so far. Before building this app, I had limited experience with backend engineering, especially with relational databases and cloud deployment. Setting up PostgreSQL, writing SQL joins and migrations, and integrating the database with FastAPI through asynchronous endpoints introduced countless errors that I had to troubleshoot one-by-one. Designing schemas that could handle user profiles, scholarship metadata, and vector embeddings while maintaining normalization and performance took weeks of iteration. I also had to learn how to structure data relationships so that the system could scale without breaking under multiple concurrent requests.
The integration with Pinecone was another major challenge. Vector databases were relatively new to me, and understanding how embeddings, cosine similarity, and upserts worked in production required a deep dive into technical documentation. I encountered several failures while indexing large batches of embeddings, often due to small mistakes in data formatting or inconsistencies in the API calls. Getting Pinecone and PostgreSQL to communicate seamlessly while maintaining synchronization between structured and unstructured data was one of the hardest parts of development.
The algorithm itself was also far more advanced than anything I had built before. The full pipeline included a Sentence-BERT embedding model for semantic encoding, a hybrid ranking system that mixed rule-based eligibility filters with collaborative filtering, and a retrieval layer that merged vector similarity with keyword search. Each component essentially introduced its own debugging process. Training and testing the models locally, optimizing the inference time, and tuning weights for ranking relevance pushed my understanding of machine learning systems beyond what I had learned before.
Despite all of these difficulties, I am genuinely grateful for every one of them. I kept detailed notes of every error and solution, which helped me document the project and recognize my own progress. As a result, I believe I have grown immensely as both a programmer and a problem solver.
Accomplishments that we're proud of
Built a fully functional, cross-platform iOS and Android app from scratch with a production-ready full-stack architecture.
Structured and curated a dataset of 10,000+ verified scholarships across 30+ disciplines.
Designed and implemented a hybrid ranking engine combining eligibility filtering, weighted keyword matching, and collaborative signals.
Integrated semantic search using SBERT embeddings and vector similarity indexing.
Deployed a Retrieval-Augmented Generation (RAG) pipeline with a fine-tuned FLAN-T5 model for AI-powered application assistance.
Achieved real-time personalized scholarship feeds with scalable backend infrastructure.
Delivered a polished, modern UI with secure profile management and dynamic updates.
What we learned
Building Scholify from concept to completion forced me to think like both an engineer and also as a user. I learned how to break a large, abstract idea into smaller, concrete steps, then piece those steps together into a functioning product that could make a real impact. Beyond coding, I learned the importance of persistence. Nearly every stage of this project involved a roadblock that I had to overcome, from debugging database schemas to optimizing model inference. I started documenting every problem I faced, which helped me develop a stronger, more systematic way of thinking. At one point, I also learned the value of stepping back from the developer’s perspective and viewing the app through the eyes of a student. I started asking myself, “Would I enjoy using this if I were the user?” That mindset shift changed everything. It helped me simplify complex features, redesign confusing interfaces, and focus on clarity instead of complexity. I’ve come to understand that truly great software combines solid engineering with thoughtful design. This challenge also gave me confidence in sharing my work publicly. Explaining Scholify to others showed me that the real test of understanding is the ability to distill complex ideas into a clear and relevant presentation. When I spoke about how the app worked, I learned to focus less on jargon and more on the problem it solved and the people it could help. That shift helped me grow as both a communicator and a developer. I began to see that innovation only matters when others can see its value and purpose. Presenting Scholify taught me to connect my technical work with its human impact, and that lesson will stay with me long after this competition. Overall, the experience taught me how to combine creativity, rigor, and empathy to turn an idea into something tangible. It strengthened my belief that technology, when built with care, can be a real force for opportunity and equity.
What's next for Scholify
If I were to build Scholify 2.0, my focus would be on making the system more intelligent, personalized, and adaptive through the integration of agentic workflows. The current version already uses AI models and a hybrid ranking engine to match scholarships effectively, but the next generation of the app would transform that intelligence into something far more interactive and human-like in its reasoning.
In Scholify 2.0, every student would have their own dedicated AI Agent, trained dynamically on their profile data, goals, and past activity. Instead of a single centralized recommendation engine, these agents would act as personalized copilots that can reason, plan, and execute actions on behalf of the user. For example, an agent could identify a newly added scholarship that fits a student’s evolving interests, automatically draft a shortlist based on eligibility confidence scores, or remind the student of upcoming deadlines. Each agent would maintain its own memory layer, allowing it to learn from how the student interacts with the app over time.
Technically, this would involve building an orchestrated multi-agent workflow that combines retrieval, reasoning, and generation into a continuous feedback loop. The agents would rely on vectorized scholarship data from Pinecone, semantic search pipelines for dynamic retrieval, and an LLM-based reasoning module that can autonomously call functions, analyze results, and plan next steps. This system would create an active discovery experience, augmenting the current recommendation feed.
Beyond intelligence, I would also focus on infrastructure and transparency. Scholify 2.0 would adopt a modular microservice architecture where each component, including the vector store, recommendation pipeline, and agentic layer, would run as an independent service. This would make the system more scalable, maintainable, and cloud-efficient. I would also implement explainability features that show students exactly why each scholarship was suggested, helping them trust and understand the algorithm’s reasoning.
Lastly, I would introduce new community and mentorship features that build on the same foundation. Students could connect with peers applying for similar scholarships or with mentors who have already gone through the process. I would also create a space for verified sponsors to directly publish their scholarships on Scholify. This would let organizations use the app as a bridge between those offering support and those seeking it. As I continue to develop Scholify, I will certainly focus on building this complete and collaborative ecosystem for both sides.
Log in or sign up for Devpost to join the conversation.