Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

58 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Dunong — AI Recitation Coach

Oral confidence coaching for Filipino students. Scan a printed module, explain it out loud, and get AI feedback that builds your confidence to speak — not just your grade.

Hackathon: ACM TechSprint: Asteria · Organized by FIT ACM Student Chapter · Major partner: Accenture · Tech partner: GitHub Project Case: Case 2 — AI-Powered Study Companion for Filipino Learners


1. ReadMe

Dunong narrows in on one specific, underserved gap: oral confidence and recitation anxiety among Filipino high school and college students. Instead of a generic Q&A chatbot, it helps students find their voice out loud — in a private, judgment-free space — before they have to perform in a real classroom.

How it works:

  1. Scan — Point the camera at a printed module or upload a photo. AI extracts the text and shows a "Tama Ba?" step so the student can correct any misread words before continuing.
  2. Lock ("Sipat-Aral") — Tap the paragraph to recite. Only that text is used as context.
  3. Recite ("Sabihin Mo") — Explain the locked text out loud in Filipino, English, or code-switched. Difficulty climbs through a confidence ladder: Read-Aloud → Paraphrase → Cold Call → Stand & Deliver.
  4. Feedback — Scored on accuracy, clarity, and confidence (filler words, hesitation), with specific and encouraging AI-generated coaching.
  5. Chat — Ask Dunong anything about the subject. Works online and falls back to offline keyword responses when no connection is available.

Only the OCR'd text and speech transcript — a few kilobytes — ever touch the network, keeping the app genuinely low-bandwidth and mobile-first.


2. Setup Guide

Prerequisites

Backend

cd backend
npm install
cp .env.example .env    # paste your GROQ_API_KEY into .env
npm run dev             # starts on http://localhost:3001

Leave GROQ_API_KEY blank to run in mock mode — the app still works with canned responses, no API key required.

Frontend

cd frontend
npm install
npm run dev             # starts on http://localhost:5173

The frontend proxies /api/* to localhost:3001 automatically in development.

Environment Variables

File Variable Description
backend/.env GROQ_API_KEY Groq API key for AI features
backend/.env PORT Backend port (default: 3001)
frontend/.env VITE_API_URL Backend URL for deployed builds (leave empty for local dev)

Deployed Version


3. Tech Stack

Layer Technology
Frontend React 18, TypeScript, Vite, Tailwind CSS
Backend Node.js, Express, TypeScript, tsx
AI — Vision (OCR) Groq API · meta-llama/llama-4-scout-17b-16e-instruct
AI — Coaching & Chat Groq API · llama-3.3-70b-versatile
AI — Speech-to-Text Groq Whisper · whisper-large-v3-turbo
Text-to-Speech Web Speech API (on-device, no cost)
Deployment Render (backend), Vercel/Netlify (frontend)

Architecture:

Browser / Phone
  Camera / Gallery → canvas → JPEG (max 1280px) → POST /api/scan
  Mic → MediaRecorder (webm/mp4) → POST /api/transcribe
  Text → POST /api/coach → accuracy + confidence + clarity scores
                ↓
  Express backend (Render)
    /api/scan       → Groq Vision (Llama 4 Scout) → extracted text
    /api/transcribe → Groq Whisper → transcript
    /api/coach      → Groq LLM (Llama 3.3 70B) → feedback + scores
    /api/chat       → Groq LLM → study companion response
    /api/ask        → Groq LLM → paragraph explanation

4. AI Disclosure

AI tools used to build Dunong

Tool How it was used
Claude (Anthropic) Primary coding assistant — UI implementation, bug fixes, architecture decisions
ChatGPT (OpenAI) Generated the mascot PNG images used in the app
Groq Console Testing and validating AI model responses during development

AI embedded inside the product

Feature Model Provider Purpose
Scan Module (OCR) Llama 4 Scout 17B Groq Extract printed text from a photo
Coaching Feedback Llama 3.3 70B Groq Score recitation and generate personalized feedback in Filipino/English
Speech Transcription Whisper Large V3 Turbo Groq Convert student's spoken answer to text
AI Chat Llama 3.3 70B Groq Answer student questions about any subject
Ask Dunong Llama 3.3 70B Groq Explain a selected paragraph in Filipino or English
Text-to-Speech Web Speech API On-device Read prompts and feedback aloud (no API cost, no data sent)
Offline fallback Keyword rules None Basic responses when there is no internet connection

Privacy: No images are stored. No audio is stored. Only the OCR-extracted text and speech transcript (plain text, a few KB each) are sent to the Groq API for processing. Nothing is persisted beyond the current session.


5. Team Members

Team Name: Fine Tuned

Name Role
Shintaro Suzuki Demo Video
Melfred Bernabe Full Stack
Kenth Razen Magbanua Dev Ops
Shana Czane M. Cruzat Representative

About

Dunong aims to empower and build the oral confidence of Filipino students. Scanning printed modules via VLM, it uses a bilingual voice loop to listen as you explain concepts out loud. Practice in a judgment-free space, track filler words, build streaks, and conquer classroom anxiety before you raise your hand. Hybrid offline.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages