Mystagogue Ai
MYSTAGOGUE AI — Making Global Education Accessible Through Intelligent Automation Target Audience International students, education agencies, and future university partners. The Problem International admissions are fragmented, repetitive, and confusing, requiring students and agencies to re-enter data across dozens of disconnected portals. Our Solution: Universal Student Passport Mystagogue AI unifies admissions around a single reusable student profile: Discovery: Program search with verified sources. Eligibility Engine: Deterministic rule-based evaluation (no AI guessing). AI Document Prep: Intelligent extraction with mandatory student oversight. ApplyOnce Workspace: Consent-controlled application drafts. Finance: Tuition, living costs, and scholarship comparison. Tracking: Dashboard with agency collaboration and role-based access. Why AI & Hybrid Architecture AI extracts data and parses complex requirements, while deterministic software strictly controls eligibility, approvals, and application states. Prototype & Tech Stack Built with Next.js, NestJS, PostgreSQL, and FastAPI/Python. Core workflows, security guards, and sandbox submissions are fully tested. Competitive Advantage Connects profile, discovery, eligibility, prep, consent, and tracking into one platform. Prioritizes student data ownership and realistic integration boundaries over generic chat. Vision A scalable SaaS removing administrative barriers between global talent and higher education.
Preliminary score
- Value for the user14.5 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI11.5 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing8.0 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility7.0 / 15
Data requirements, running costs and a clear next step.
- Originality5.0 / 10
A meaningful difference from common solutions.
What the judges wrote
- Claude Opus 5.5Preliminary43/100
- Strengths
- Your problem framing is honest and disciplined: the deck says explicitly that no time, error or conversion savings are claimed, and that market sizes are formulas only, which is unusual and credible. Your hybrid design stands out. The LLM only proposes extracted fields as 'suggestions', while eligibility runs on deterministic rules and every approved application is frozen with a SHA-256 snapshot digest. Your testing write-up names real engineering failures you found and fixed, such as markdown code blocks in AI output, PII leaking into AI logs, race conditions and digest serialization mismatches.
- To improve
- Your deck labels every slide 'Proposed' and never shows the working product, so judges cannot see the core scenario run end to end. Add screenshots or a short walkthrough going from document upload to extracted fields to eligibility result to approved draft. Your testing covers infrastructure but not AI quality: measure extraction accuracy on a set of real or synthetic transcripts and diplomas, list the fields it got wrong, and compare the time against filling forms manually. Your track asks for one metric that justifies adoption, so name one, add a rough estimate of LLM and hosting cost per student, and replace placeholders like [PILOT RESULT] with a concrete next step.
- GPT-6 AstraPreliminary49/100
- Strengths
- Your reusable Student Passport targets a concrete burden for students and agencies: repeatedly entering academic information across disconnected portals. Your proposed separation of AI extraction from deterministic eligibility, alongside approval snapshots and manual fallbacks, is appropriate for high-stakes applications. Your failure report identifies specific engineering problems, including serialization mismatches, concurrent requests, and PII leakage in logs.
- To improve
- Your materials should distinguish implemented features from the deck's proposed architecture and demonstrate one document-to-eligibility-to-approved-draft sandbox flow, including actual AI inputs, outputs, and a receipt. Support your testing report with reproducible cases, expected versus actual results, and execution counts; the current evidence does not establish production readiness. Measure agency handling time and correction rates against manual processing on the same applications. Identify an initial program catalog, its sources and maintenance process, and estimated per-application AI, hosting, and human-verification costs.
The team
Eshgin AlyarovLead
developer