Inspiration
Every student hunting for scholarships, fellowships, internships or grants runs into the same wall: the information is scattered across dozens of sites, deadlines and requirements are buried in PDFs, and scam or cloned offers look just as polished as real ones. Students lose hours, and sometimes money or personal data, before they ever submit an application. We wanted one place where a student can ask "Is this real? Am I eligible? What do I do next?" and get an answer they can actually trust, in English or Bangla.
What it does
ProofPath AI is a student opportunity intelligence, verification and protection platform.
- Verification Workspace: submit a URL, pasted text, PDF or screenshot. The AI extracts the organization and its claims, checks them against primary sources, and returns a verdict (Verified, Supported, Contradicted, Unverified, Insufficient Evidence or Expired) with cited evidence.
- Evidence Relationship Explorer: an interactive graph tracing Submission → Organization → Claims → Evidence → Verdict, plus an audit history you can compare side by side.
- Deterministic Eligibility Engine: rule-based matching of GPA, English proficiency (IELTS / TOEFL / Duolingo), deadline and profile data. Outcomes are Eligible, Potentially Eligible, Not Eligible or Missing Info.
- Opportunity Intelligence Score (0–100): a transparent score made of four drivers:
$$S = P_{\le 30} + M_{\le 30} + R_{\le 25} + H_{\le 15}$$
where $P$ is profile completeness, $M$ verified match coverage, $R$ application readiness and $H$ verification hygiene.
- Application Intelligence: kanban, list and monthly calendar views with readiness checklists for missing documents.
- Bilingual AI Advisor: saved conversations that use the student's profile as context.
- Admin & Moderation: role-based panel, community scam reports, audit logs and AI usage telemetry.
How we built it
- Frontend: React 19, TypeScript, Tailwind CSS v4, Lucide icons and Motion, with light and dark themes and client-side PDF export via jsPDF.
- Backend: Express + TypeScript with Zod validation, HTTP-only cookie sessions, scrypt password hashing, CSRF tokens, sliding-window rate limiting and role-based access control.
- Database: SQLite (better-sqlite3) in WAL mode with foreign keys and automatic migrations across 20 normalized tables.
- AI engine: server-side Google Gemini with Google Search grounding, URL context inspection and multimodal PDF/image analysis, so API keys never touch the browser.
- Quality: a Vitest integration and security test suite plus strict TypeScript checks.
Challenges we ran into
- Keeping AI honest: we deliberately split the work. AI handles open-ended verification, while eligibility and scoring are computed by deterministic code so results are explainable and repeatable.
- Safe URL fetching: letting users submit links is an SSRF risk, so we built guardrails that block localhost, private ranges and cloud metadata addresses before any request.
- Safe uploads: we validate file type by MIME allowlist and magic bytes, not just file extension.
- Making evidence understandable: turning claims, sources and verdicts into a graph a student can read at a glance took several design iterations.
- Designing one coherent product across verification, eligibility, tracking and chat without it feeling like separate tools.
Accomplishments that we're proud of
- A true full-stack product with auth, a real database, an AI engine and an admin side, not just a front-end demo.
- Security treated as a feature: CSRF, SSRF protection, upload validation, rate limiting and audit logging.
- A scoring and eligibility system where every number can be explained to the student.
- English and Bangla support, so the tool reaches students who are often underserved by English-only platforms.
What we learned
- Grounding AI answers in retrieved sources makes them far more trustworthy than free-form generation.
- Combining AI with deterministic rules gives the best mix of flexibility and reliability.
- Security and data modeling decisions are much cheaper to make early than to retrofit.
- Small UX details, like readiness checklists and plain-language verdicts, matter as much as the AI itself.
What's next for ProofPath AI
- Automated opportunity discovery and deadline alerts by email and push.
- A browser extension that flags suspicious scholarship pages as students browse.
- More languages beyond English and Bangla.
- Document-aware application help, such as checking essays and CVs against each opportunity's requirements.
- Partnerships with universities and verified scholarship providers to build a trusted opportunity registry.
Built With
- better-sqlite3
- express.js
- gemini-api
- google-gemini
- google-search-grounding
- jspdf
- lucide-react
- motion
- multer
- node.js
- react
- rest-api
- sqlite
- tailwindcss
- typescript
- vite
- vitest
- zod