Skip to content

amansethhh/Orato-AI

BranchesTags

Repository files navigation

Orato AI

Real-time AI-powered communication coach that analyses spoken responses and delivers structured, actionable feedback.

Voice Analysis • AI Feedback • Interview Coaching • Production-Ready

⭐ If you like this project, consider giving it a star!

🎬 Live Demo

Orato AI Demo

Real-time voice recording • AI-powered feedback • Structured scoring • Sample answers • Communication coaching

📋 Overview

The Problem

Most people struggle with spoken communication — interviews, presentations, and everyday conversations — yet receive little to no structured feedback on how they actually sound. Traditional coaching is expensive, inaccessible, and not available in real time.

The Solution

Orato AI is a production-grade, full-stack AI coaching platform. Users record their spoken responses, and the system instantly transcribes, analyzes, and returns structured coaching feedback — scoring clarity, confidence, fluency, and structure — powered by Google Gemini or OpenAI.

Key value: Real-time, structured, actionable coaching available to anyone with a browser.

🖼️ UI Preview

Home / Mode Selection Voice Practice Session AI Feedback Results
Home screen showing practice mode selection Voice recording interface with live waveform Structured AI feedback with scores and coaching tips
Choose from Interview, Presentation, or Casual Speaking modes Record your response with live audio waveform visualization Receive structured scores, strengths, improvements, and a sample answer

✨ Features

Feature Description
🎙 Real-time voice recording Capture spoken responses using the MediaRecorder API with live waveform visualization
🧠 AI-powered feedback Coaching analysis via Google Gemini API or OpenAI (configurable)
📊 Structured scoring Scores across Clarity, Confidence, Fluency, and Structure (0–100)
🗣 Speech-to-text Browser-native Web Speech API transcription with no third-party cost
🔊 Text-to-speech playback AI feedback narrated back via SpeechSynthesis API
📈 Session analytics Track improvement over time with per-session history and trend charts
🌐 Multilingual support Full English and Hindi interface via built-in translation layer
⚡ Adaptive question generation AI-generated questions tailored to role, difficulty, and session history
🛡 Secure backend proxy API keys never exposed to the browser — all AI calls routed through the Express server
💾 In-memory caching Response caching and transcript deduplication to reduce API cost and latency
⚠️ Graceful fallback Intelligent fallback feedback when AI is unavailable — no broken states
🌙 Dark mode Full dark/light theme support

🏗 Architecture

+------------------------------------------------------------+
|                  FRONTEND (React + Vite)                   |
|                                                            |
|  [Home] -> [Setup] -> [VoicePractice] -> [AIFeedback]      |
|  MediaRecorder API          Web Speech API (STT)           |
|  SpeechSynthesis API (TTS)  Recharts (analytics)           |
+-----------------------------+------------------------------+
                              |
                   POST /api/ai (transcript + prompt)
                              |
                              v
+------------------------------------------------------------+
|                BACKEND (Node.js + Express)                 |
|                                                            |
|  Rate Limiter -> Request Validator -> Cache Lookup         |
|                         |                                  |
|             +-----------+-----------+                      |
|             v                       v                      |
|      Google Gemini API        OpenAI API                   |
|             +-----------+-----------+                      |
|                         v                                  |
|  Response Validator -> Cache Store -> JSON Response        |
+-----------------------------+------------------------------+
                              |
                   Structured feedback JSON
                              |
                              v
                     +----------------+
                     |   React UI     |
                     |   Score cards  |
                     |   Coaching     |
                     |   insights     |
                     +----------------+

🔁 Request Flow

User speaks -> MediaRecorder captures audio
           -> Web Speech API transcribes to text
           -> Frontend sends transcript + prompt to backend
           -> Backend checks cache (dedup + prompt cache)
           -> Backend calls Gemini / OpenAI with schema prompt
           -> Response validated, clamped, cached
           -> Structured JSON returned to frontend
           -> UI renders scores, strengths, improvements, sample answer
           -> SpeechSynthesis reads feedback aloud (optional)

⚙️ Tech Stack

Frontend

Technology Version Purpose
React 18 UI framework
Vite 6 Build tool & dev server
Tailwind CSS 3 Utility-first styling
Framer Motion 11 Animations
Recharts 2 Session analytics charts
React Router 6 Client-side routing
React Query 5 Server state management

Backend

Technology Version Purpose
Node.js 18+ Runtime
Express 4 HTTP server
express-rate-limit 7 API rate limiting
dotenv 16 Environment config
cors 2 Cross-origin policy

AI & Browser APIs

API Purpose
Google Gemini API Primary AI provider for feedback & question generation
OpenAI API Optional secondary AI provider
Web Speech API (SpeechRecognition) Browser-native speech-to-text
MediaRecorder API Audio capture with waveform
SpeechSynthesis API Text-to-speech feedback playback

🔄 How It Works

 1. User selects a practice mode (Interview / Presentation / Casual)
        |
 2. AI generates an adaptive question based on role and difficulty
        |
 3. User records their spoken response via the microphone
        |
 4. Web Speech API transcribes the audio to text in real time
        |
 5. Transcript and coaching prompt are sent to the backend proxy
        |
 6. Backend checks the in-memory cache; on a miss, calls Gemini or OpenAI
        |
 7. AI returns structured JSON: scores, strengths, improvements, tip, sample answer
        |
 8. Frontend renders animated score cards, coaching insights, and a model answer
        |
 9. SpeechSynthesis reads the feedback aloud if enabled
        |
10. Session is saved; analytics charts update with the new data point

📂 Project Structure

Orato-AI/
├── index.html
├── vite.config.js
├── tailwind.config.js
├── package.json
│
├── src/
│   ├── main.jsx                # App entry point
│   ├── App.jsx                 # Route definitions
│   ├── Layout.jsx              # Shell layout
│   │
│   ├── pages/
│   │   ├── Home.jsx            # Mode selection dashboard
│   │   ├── Intro.jsx           # Onboarding / landing
│   │   ├── InterviewSetup.jsx  # Configure interview parameters
│   │   ├── QuestionSetup.jsx   # Custom question setup
│   │   ├── VoicePractice.jsx   # Recording + STT interface
│   │   └── AIFeedback.jsx      # Feedback results & analytics
│   │
│   ├── components/
│   │   ├── VoiceWaveform.jsx   # Live audio visualizer
│   │   ├── ScoreIndicator.jsx  # Animated score ring
│   │   ├── SessionSummary.jsx  # Per-session summary card
│   │   ├── ProgressSnapshot.jsx
│   │   ├── SettingsModal.jsx
│   │   ├── SettingsProvider.jsx
│   │   ├── ErrorBoundary.jsx
│   │   ├── translations.jsx    # EN / HI strings
│   │   └── ui/                 # Radix UI + shadcn components
│   │
│   ├── api/
│   │   ├── apiClient.js        # Fetch wrapper (frontend -> backend)
│   │   ├── aiService.js        # AI endpoint helpers
│   │   └── sessionAnalytics.js
│   │
│   ├── lib/
│   │   ├── AuthContext.jsx
│   │   ├── logger.js
│   │   ├── healthCheck.js
│   │   └── utils.js
│   │
│   └── utils/
│       └── index.js
│
└── server/
    ├── index.js                # Express server -- AI proxy, cache, rate limit
    ├── config.js               # Centralised server config
    ├── package.json
    └── .env.example            # Environment variable template

🛠 Setup Instructions

Prerequisites

  • Node.js 18+
  • A Google Gemini API key or an OpenAI API key

1. Clone the repository

git clone https://github.com/amansethhh/Orato-AI.git
cd Orato-AI

2. Install frontend dependencies

npm install

3. Install backend dependencies

cd server
npm install
cd ..

4. Configure environment variables

cp server/.env.example server/.env

Open server/.env and add your API key:

# Choose your AI provider: auto | gemini | openai
AI_PROVIDER=auto

GEMINI_API_KEY=your_gemini_key_here
OPENAI_API_KEY=your_openai_key_here   # optional

PORT=3001
FRONTEND_URL=http://localhost:5173
NODE_ENV=development

Note: If neither key is provided, the server runs in mock mode and returns locally-generated coaching feedback — useful for UI development without an API key.

5. Start the backend

cd server
node index.js

Expected output:

======================================================
           Orato AI - Backend Server
======================================================

[INFO] Server running on http://localhost:3001
[INFO] AI Provider: GEMINI

6. Start the frontend

Open a new terminal from the project root:

npm run dev

The app opens automatically at http://localhost:5173.

🔐 Environment Variables

Backend — server/.env

Variable Required Default Description
GEMINI_API_KEY Yes* — Google Gemini API key
OPENAI_API_KEY Yes* — OpenAI API key
AI_PROVIDER No auto Force a provider: auto, gemini, openai
PORT No 3001 Backend server port
FRONTEND_URL No http://localhost:5173 Allowed CORS origin
NODE_ENV No development development or production

*At least one AI key is required for real feedback. Without either, the server runs in mock mode.

Frontend — .env (optional)

Variable Description
VITE_API_URL Backend base URL (defaults to /api via Vite proxy)

⚡ Performance & Optimization

Optimization Implementation
Code splitting Vite manualChunks separates React, UI libs, and charts into parallel-loaded bundles
Lazy loading Route-level code splitting via React Router lazy imports
In-memory caching Backend caches AI responses by prompt hash (TTL: configurable)
Transcript deduplication Identical transcripts return cached results within a dedup window
Rate limiting express-rate-limit prevents abuse and protects API quota
Request validation Prompt and transcript length limits enforced server-side before any AI call
Exponential backoff Automatic retry with backoff on AI rate-limit (429) responses
Graceful fallback Context-aware locally-generated feedback ensures zero broken states
Timeout handling Per-request AbortController timeout on all AI calls

📡 API Reference

GET /api/health

Returns server status, active AI provider, uptime, and cache size.

{
  "status": "ok",
  "provider": "gemini",
  "uptime": 142,
  "cacheSize": 3,
  "timestamp": "2025-01-01T00:00:00.000Z"
}

POST /api/ai

Analyze a spoken response and return structured coaching feedback.

Request body:

{
  "prompt": "The user answered the interview question: 'Tell me about yourself'",
  "transcript": "Hi, I'm a software engineer with 3 years of experience..."
}

Response:

{
  "feedback": "Your opening established context well. Consider adding a specific achievement to strengthen impact.",
  "clarity": 78,
  "confidence": 65,
  "structure": 72,
  "fluency": 80,
  "strengths": ["Clear introduction", "Good pacing"],
  "improvements": ["Add a quantifiable result", "Reduce filler words"],
  "tip": "Use the STAR framework: Situation, Task, Action, Result.",
  "sampleAnswer": "I'm a software engineer with 3 years of experience...",
  "filler_word_count": 4,
  "_provider": "gemini"
}

POST /api/ai/question

Generate an adaptive practice question.

Request body:

{
  "prompt": "Generate a behavioral interview question for a senior frontend engineer role, medium difficulty."
}

Response:

{
  "question": "Describe a situation where you had to refactor a large codebase under time pressure. How did you prioritize?"
}

🚧 Future Improvements

  • Real-time streaming AI — Stream AI tokens to the UI for lower perceived latency
  • User authentication — Persistent accounts with cross-device session history
  • Cloud deployment — Docker + Railway / Render backend, Vercel frontend
  • Whisper STT — Server-side transcription via OpenAI Whisper for higher accuracy
  • Video analysis — Webcam-based posture and eye-contact scoring
  • Interview simulation — Multi-turn conversational interview mode
  • PDF export — Download session feedback reports (html2canvas + jsPDF already included)
  • Team / recruiter dashboard — Share sessions and track candidate progress

🤝 Contributing

Contributions are welcome. Please follow these steps:

  1. Fork the repository
  2. Create a feature branch: git checkout -b feature/your-feature-name
  3. Make your changes and commit: git commit -m "feat: add your feature"
  4. Push to your fork: git push origin feature/your-feature-name
  5. Open a pull request against main

Please keep pull requests focused and include a clear description of the change and its motivation.

📄 License

This project is licensed under the MIT License.

Built with ❤️ by amansethhh

About

Orato AI is a privacy-first web app for improving spoken communication through real-time voice analysis. It provides structured practice modes, AI-generated feedback on delivery, and adaptive coaching without storing audio or compromising user privacy.

oratoai.base44.app

Topics

aicommunication-coachexpressfullstackgemini-apinodejsopenai-apireactspeech-recognitiontailwindcsstext-to-speechvitevoice-aiweb-app

Resources

Readme
MIT license
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages