AI Enterprise SolutionsGPT & Claude

Syntrio

Students today can use AI to solve homework in seconds, but getting the right answer does not always mean understanding the topic. Many students copy solutions without learning from their mistakes. The main problems are: 1. Lack of critical thinking. Students may rely on AI-generated answers instead of developing their own problem-solving skills. 2. Hidden learning gaps. Students, parents, and teachers often struggle to identify why mistakes happen and which topics need improvement. 3. Different learning needs. Students from grades 1 to 11 have different learning levels, and children with dyslexia or visual and hearing impairments may need more accessible explanations. The core issue: Existing tools often focus on providing answers or tracking grades rather than helping students understand their mistakes. Students need guidance to discover errors independently, while parents and teachers need clear insights into their progress. Our Solution: SokratikAI SokratikAI is an AI-powered learning platform that helps students learn through questions instead of ready-made answers. It analyzes each solution, identifies mistakes, and provides a score from 0 to 100. Rather than revealing the correct answer, it asks guiding questions and offers step-by-step hints, helping students understand their mistakes and solve problems independently.

Open the demo GitHub

Preliminary score

58.5/ 100Promisingaverage of 2 judges

Value for the user16.5 / 25
A specific problem and a clear outcome from solving it.
Prototype and use of AI18.5 / 30
A working core scenario, and what AI actually contributes to the result.
Quality testing11.5 / 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

GPT-6 AstraPreliminary61/100
Strengths
Your focus on diagnosing mistakes rather than supplying answers addresses a concrete need for students, parents, and teachers. Your screenshots show child profiles, topic/error reports, and an AI-generated parent response that acknowledges missing practice data. Your testing notes provide specific integration failures and fixes while clearly disclosing synthetic data, provider limits, and unvalidated curriculum support.
To improve
Your next demonstration should trace one incorrect solution through guiding questions, a saved attempt, and updated parent and teacher views; resolve the unfinished teacher-dashboard test entry. Your evaluation should use teacher-labelled examples to measure diagnostic accuracy, premature answer disclosure, and review-time savings against current practice. Your deployment documentation should reconcile Gemini in the submission with OpenAI/Claude in the README and specify running costs, quota behaviour, and protections for children's records. Your next milestone should be a consent-based pilot limited to one grade and topic, including accessibility checks with intended users.
Claude Opus 5.5Preliminary56/100
Strengths
Your core idea of an AI that asks guiding questions instead of giving answers, and records why a student erred (concept, calculation, prerequisite gap), is well grounded. The Aysel and Mr. Rashad personas make the problem concrete. The deck screenshots show a working Azerbaijani-language app: parent and teacher roles, child profiles with special-needs settings, per-topic mistake rates and Gemini-generated advice for parents. Your testing section is refreshingly honest: it lists real breakages (provider credits, env keys, a frontend/backend contract mismatch), says what you did not test, and makes no invented numeric claims.
To improve
Your Socratic chat is the heart of the product, but no screenshot or transcript shows it working. Run a small set of test problems with known wrong solutions and report how often the tutor leaks the answer and how accurately it classifies the error type. Item 6 of your testing write-up was left as an unfilled template placeholder, and the README still references OpenAI/Claude and a different demo teacher account, so tidy these before the final. Add running costs per student, explain the bracketed 42/65 AZN prices, and name one concrete next step, such as a pilot with one 5th-grade class and a metric like teacher review time or repeat-mistake rate.

The team

Eyvazov Kamran
developer
Əhmədov Nicat
product
Məmmədli ƏdiləLead
developer

Disclosure

Required by the rules. Judges verify it against the repository.

Models: gemini-3.5-flash,gemini-2.5-flash,gemini-3.1-flash-lite,gemini-3.5-flash-lite,gemini-2.5-flash-lite,gemini-2.5-pro
Data: Synthetic only. Demo school class (12 fictional students, 180 simulated practice attempts) generated by our seed script. No real student data and no public datasets. Additional records are created live by users during the demo (chat attempts, error types) and stored in SQLite.
Components: Node.js, Express, SQLite (better-sqlite3), vanilla HTML/CSS/JS, Gemini API (OpenAI-compatible endpoint), Web Speech API
Built during the event: Confirmed

Quality testing

WHAT WE TESTED We manually tested the full flow end to end on a demo class (12 fictional students, simulated attempts): parent registration/login, adding a child with a class code, child mode (Socratic chat, Simplify, Read aloud, large font), parent report with AI advice, and the teacher dashboard with the AI class summary. We checked that chat attempts are saved to SQLite and appear in the parent report and class statistics. We also tested role separation (a parent cannot open teacher data) and the multi-model fallback by letting it skip unavailable models. WHAT BROKE (and how we fixed it) 1. The AI provider ran out of credits mid-build ("no credits remaining"). We rebuilt the AI layer as provider-agnostic and added automatic fallback across several Gemini models, with a cooldown for models that return 404/429/503. 2. The AI returned "API key undefined": the key was in .env.example instead of .env. Fixed, and we removed the key from the example file. 3. A teammate's machine showed "wrong password": SQLite files are not committed to git, so his database was empty. We now seed demo data with a script. 4. Our new frontend and backend did not match: it used usernames and a "student" role, while the backend uses email and parent/teacher roles. We aligned the login contract. We also dropped the separate student account on purpose: young children have no phones, so the child works inside the parent's account ("child mode"). 5. After splitting the login page from the app, CSS and logout redirects broke. We separated stylesheets and fixed redirects. 6. [WRITE HONESTLY: e.g. "The teacher dashboard still failed to render with the new frontend at submission time" OR "fixed by ..."] WHAT WE DID NOT TEST No real students, teachers or classroom pilot. No load testing. Free-tier AI limits can cause slow or failed replies under heavy use. AI answers are not verified against the national curriculum, and Azerbaijani output quality was only checked by us informally. Voice reading depends on browser voices (it may fall back to a Turkish voice). COMPARISON WITH TODAY Today a teacher has to review notebooks manually to find out who struggles with what, and parents usually learn about gaps only after a bad grade. General chatbots give students the final answer. Our tool asks guiding questions instead, records why the student erred (concept, calculation, or prerequisite gap), and shows teachers and parents the pattern automatically. We have not measured learning outcomes or time saved, so we make no numeric claims.

Worth 20 of the 100 points.

Running it

Parents: Email: [email protected] Password: demo1234 Teachers: Email: [email protected] Password: demo1234

Project ownership

The team keeps the rights to this project. Judges were given access for verification only.