MindBridge
Target User: Students and early-career professionals seeking personalized skill development and career mentorship. The Problem: Learners struggle with fragmented educational resources and generic, one-size-fits-all guidance. They waste time trying to find the right courses and mentors that match their specific skill gaps, leading to low engagement, lack of real-time feedback, and delayed career growth.
Preliminary score
- Value for the user14.5 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI17.5 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing8.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility8.5 / 15
Data requirements, running costs and a clear next step.
- Originality4.5 / 10
A meaningful difference from common solutions.
What the judges wrote
- Claude Opus 5.5Preliminary47/100
- Strengths
- Your README is unusually honest and specific about demo boundaries: it says plainly what is real and what is simulated, and your 'Working now. Still to prove.' slide does the same. The prototype covers a lot of ground: bilingual EN/AZ UI, explainable TF-IDF cohort ranking with visible score breakdowns and age eligibility, a whiteboard, and a certificate gate tied to quiz, code checks and a 70% final. Running the coach on-device with SmolLM2, keeping profile data out of prompts and labelling checked answers separately from generated ones, is a thoughtful privacy and cost choice.
- To improve
- Your materials contradict each other on the AI: the submission and deck cite live Gemini answers and Gemini rate-limit fixes, while the README says the coach is keyless and runs only a 135M local model. Clarify what actually runs. The AI is not yet essential: matching is deterministic TF-IDF plus rules over sample cohorts, and a 135M model is weak for tutoring, especially in Azerbaijani. Back your testing with specifics, such as what the 28 checks cover, sample coach answers that failed, and matching quality against a baseline or how learners find tutors today. Narrow the user (for example Baku high-schoolers preparing for a specific exam), source the '1,200 respondents' statistics, and tie the product to the Enterprise track's workflow-and-metric framing.
- GPT-6 AstraPreliminary60/100
- Strengths
- Your prototype connects explainable, age-aware cohort recommendations with a bilingual study coach, assessments, and certificate eligibility rather than stopping at a chat interface. Your README clearly distinguishes checked answers from generated responses and implemented browser behavior from simulated services. The four-input exercise checks, execution timeout, and certificate prerequisites give the learning journey concrete completion conditions.
- To improve
- Your evidence would be stronger with a reproducible end-to-end walkthrough that reconciles the deck's live Gemini claims with the README's on-device SmolLM2 architecture. Publish the 28 check results alongside English/Azerbaijani answer-quality tests, concrete failures, and a comparison with manual cohort assignment or ordinary course search. Pilot with one defined learner segment and measure matching time or learning gains; for the enterprise track, identify the organizational buyer and workflow. Extend your pricing assumptions into per-cohort operating costs, and prioritize secure authentication and verifiable parental consent before handling real learner data.
The team
1.Pənah Pənahlı
developer
Orduxan OrduxanlıLead
product
Elgun Ibrahimli
designer
Magsad Huseynli
business
İsmixan Məmmədov
business