Inspiration
ProofPath began with a real constraint: a 19-year-old university student in China needed to find opportunities that were open to their region, still active, and realistic to pursue. Listings were easy to find. Deciding which opportunity actually fit—and why—was not.
Students rarely miss opportunities because they lack ambition. They miss them because eligibility rules are scattered, readiness is difficult to judge, and deadlines turn every missing document into a crisis. Existing directories answer “what exists?” ProofPath answers “what is realistic for me today?”
What it does
ProofPath turns scholarships, internships, fellowships, hackathons, and community programs into explainable readiness decisions.
Students can:
- Discover opportunities ranked against their own evidence
- Inspect every matched requirement, blocker, and uncertainty
- Generate a deadline-aware action plan
- Update their profile and see recommendations recalculate immediately
- Use Evidence Leverage to identify the single missing proof that improves the most paths
The result is not another endless directory. It is a decision workspace that tells students what fits, what is missing, and what to do next.
How we built it
ProofPath is a React and TypeScript application powered by a deterministic matching engine. Requirements are mapped to profile evidence and scored with an explicit 78/22 requirement-and-interest heuristic. The heuristic measures readiness; it does not predict acceptance.
Eligibility uses three states: confirmed, blocked, and unknown. A failed hard requirement blocks unsafe progression, while an unknown hard requirement generates a verify-first plan. The planning engine separates eligibility, evidence work, and available time before sequencing actions against exact UTC deadlines with a fixed Asia/Shanghai display.
GSAP supports the editorial narrative, while a custom responsive CSS system keeps the product workspace clear and practical. Profile data, saved paths, and plan progress stay in browser-local storage.
Challenges
The hardest problem was trust. A high score is useless if the student cannot inspect it. We separated the matching logic from the interface, exposed every contributing signal, and avoided fabricated precision.
The second challenge was balancing a memorable competition presentation with a genuinely useful tool. The opening creates emotional clarity; the workspace then becomes calm, dense, and action-oriented.
Accomplishments
- Complete home-to-discovery-to-plan flow with no account, API key, or setup
- Explainable scoring with visible evidence, blockers, and unknowns
- Deadline-aware plans with capacity checks and progress tracking
- Evidence Leverage across an entire opportunity portfolio
- Browser-local persistence and schema validation
- Responsive keyboard-accessible interface
- 32 automated tests covering matching, eligibility, deadlines, storage, leverage, and planning
- Lighthouse: 96 mobile performance and perfect accessibility, best-practices, and SEO scores
- Clearly labeled reviewed-source snapshots and synthetic teaching records
Responsible AI
OpenAI Codex supported research, product ideation, implementation, testing, visual and accessibility review, and submission drafting. The student builder selected the problem, supplied the real eligibility context, set publication boundaries, and retains final approval.
ProofPath itself does not call an AI model at runtime. Its scores are deterministic and inspectable. Humans must still verify source freshness, interpret eligibility, review application evidence, and make every final decision.
What we learned
Access is not only about publishing more links. It is also about reducing the interpretation work required to act. Explainability can be a product feature, not merely a disclaimer.
What's next
Next we would add verified opportunity feeds with freshness indicators, bilingual extraction, calendar export, mentor review, and broader student testing.
Try it
Built With
- css
- gsap
- phosphor-icons
- react
- typescript
- vite
- vitest
Log in or sign up for Devpost to join the conversation.