Inspiration
Many students from many regions in developing countries lack access to high-quality STM tutoring. The existing AI tools often give direct answers that prevent students from learning. This app becomes a bridge that allows students to access AI to learn with a proper method. It is powered by AI and Human responses embedded in daily apps like WhatsApp and Telegram.
What it does
TutorBuddy is an intelligent app that is integrated with WhatsApp and Telegram messaging platforms.
- Socratic AI Tutoring: The AI guides step by step to improve the learning process of students.
- Multimodal Homework Check: I can accept text and photos of homework to do or homework to check after students work on it.
- Live Human Tutor Hand-Off: The live tutor is active when students need to be guided by a tutor instead of AI.
- Dynamic Multi-Language Support: It supports English and Bahasa so students can adjust their preferred language.
- Central Admin Dashboard: Powered by Cloud Run, Firestore, and a Google Cloud Storage bucket.
How we built
- AI Core: Google Gemini 3.1 Flash Lite via
google-generativeaiSDK for fast multimodal image parsing and Socratic reasoning. - Backend Microservices: Python 3.12 hosted on GCP Cloud Functions and Cloud Run.
- Messaging APIs: Meta Graph API (WhatsApp Business API) and Telegram Bot API.
- Database: Google Cloud Firestore for real-time session tracking, user quotas, and tutor routing, and a bucket for physical data storage.
- Dashboard: HTML and python
Challenges we faced
- Multimodal API Integration: Handling raw media binaries from messaging platforms (WhatsApp and Telegram) sometimes failed to read.
- Strict Socratic Guardrails: Prompt engineering to enforce concise 1–3 sentence responses, preventing the model from giving away complete final answers before the 5th interaction turn and to make the AI focus on guided responses.
Accomplishments that we're proud of
- Successfully created a hybrid system combining Socratic AI and Live Human Tutors in a single conversational flow.
- Achieving zero-leak privacy protection for both tutors and students during live routing.
- Building a reliable multimodal pipeline capable of parsing handwritten math equations directly from low-resolution photos.
What we learned
- Advanced prompt engineering for controlled, step-by-step educational dialogues.
- Efficient binary stream management in serverless environments (GCP Cloud Functions).
- Designing resilient fallback systems for multi-channel messaging platforms.
What's next for TutorBuddy
- Expanding payment gateway integrations (Virtual Accounts & Credit Cards).
- Adding voice note parsing for voice-based Socratic tutoring.
- Scaling the Live Tutor marketplace to connect more verified STEM educators across the region.
Log in or sign up for Devpost to join the conversation.