- 1. Overview
- 2. Features
- 3. Setup Instructions
- 4. Environment Setup
- 5. File Structure
- 6. Usage
- 7. Run the Application
A voice-driven ticket counter assistant built with Streamlit, Whisper ONNX, and LLaMA integration. It helps users inquire about ticketing details through speech, providing transcription, route insights, fare estimation, and QR-based ticket metadata for reference. The assistant also offers contextual destination insights, supports admin visibility through an editable dashboard, and generates ticket summaries for operational tracking and user assistance
- 🎙️ Voice Recording: Capture speech using selected microphone devices.
- 🧠 Whisper Transcription: Convert speech to text using ONNX Whisper models.
- 🧾 LLaMA Extraction: Extract source, destination, and ticket count from transcribed text.
- 🏙️ City List: Displays supported cities from a SQLite database.
- 💰 Fare Estimation: Calculates fare and platform based on route and ticket count.
- 📦 QR Code Generation: Generates QR code with ticket metadata for reference.
- 🧭 Travel Insights: Provides contextual information about destinations using LLaMA.
- 🗃️ Ticket Logging: Stores ticket metadata and transcription in SQLite.
- 🔐 Configurable Settings: Sidebar UI for API keys, model paths, and recording options.
Before proceeding further, please ensure that all the setup steps outlined below are completed in the specified order. These instructions are essential for configuring the various tools required to successfully run the application.
Each section provides a reference to internal documentation for detailed guidance. Please follow them carefully to avoid any setup issues later in the process.
Miniconda is required to manage the application's Python environment and dependencies. Please follow the setup instructions carefully to ensure a consistent and reproducible environment.
For detailed steps, refer to the internal documentation:
Set up Miniconda
Git is required for version control and collaboration. Proper configuration ensures seamless integration with repositories and development workflows.
For detailed steps, refer to the internal documentation:
Setup Git
AnythingLLM is required as the backend server for this application. You need access to a running AnythingLLM instance to provide the LLM and RAG capabilities.
You can use either a local or remote AnythingLLM server:
- Local: Follow the AnythingLLM installation guide to set up and run the server on your machine.
- Remote: Obtain the API URL and API key from your administrator.
For detailed setup instructions, refer to the below steps
🖥️ 1. Launch AnythingLLM
- Open the AnythingLLM Desktop App or access your self-hosted instance in a browser.
🆕 2. Create a New Workspace
- Click on “New Workspace”.
- Enter a Workspace Name (e.g., test).
- Click Create.
⚙️ 3. Configure the Workspace
- Go to the Workspace Settings.
- Navigate to Agent Configuration.
- Under LLM Provider, select AnythingLLM NPU (or your preferred provider).
- Choose your model (e.g., LLaMA 3.1).
- Click Update Workspace Agent.
🔑 Step-by-Step: Get API Key and Base URL
- 🔐 4. Generate API Key
- Go to Settings → API Keys.
- Click “Generate New API Key”.
- Copy the key and store it securely.
Once your workspace is created and configured:
- Go to the Chat tab in AnythingLLM.
- Select your workspace from the dropdown.
- Type a message (e.g., “What is the capital of Karnataka?”).
- The response should come from the configured LLaMA model.
To set up the Python environment required for running the application, follow the steps below. This ensures all dependencies are installed in an isolated and reproducible environment.
-
Create your working directory:
mkdir my_working_directory cd my_working_directory -
Download and Prepare Whisper ONNX Models from Qualcomm AI-HUB:
-
This guide walks you through installing dependencies, exporting the Whisper ONNX model, and validating speech-to-text functionality on the Snapdragon X Elite platform. It includes setting up the Conda environment, installing FFmpeg, Python dependencies, QNN runtime, and downloading the Whisper-base model.
-
Follow the Run Whisper on Snapdragon X Elite guide to set up and export the whisper model from the qualcomm ai-hub-apps repository.
Setup Steps
- Step 1: Clone the repository - git clone https://github.com/quic/ai-hub-apps.git
- Step 2: Navigate to the Whisper app directory -cd ai-hub-apps/apps/windows/python/Whisper
- Step 3: Run the setup script to install all necessary dependencies.
- Step 4: Export the Whisper ONNX model using the provided script
- Step 5: Validate speech-to-text functionality using demo.py
Move Build Directory to C Drive
-
After exporting the Whisper ONNX models, you will have a folder named build in your current working directory. To make it accessible for your application,
move it to the C drive. -
Create the following repository structure to run the application.This will move the entire build directory — including the whisper_base model ONNX files, corresponding encoder/decoder folders, and model.bin — to
C:\build.🔍 Resulting Structure
C:\ └── build\ └── whisper_base\ ├── HfWhisperEncoder\ │ ├── model.onnx │ └── model.bin │ ├── HfWhisperDecoder\ │ ├── model.onnx │ └── model.bin
-
Download Your Application:
git clone -n --depth=1 --filter=tree:0 https://github.com/qualcomm/Startup-Demos.git cd Startup-Demos git sparse-checkout set --no-cone /GenAI/AI_PC/Ticket_Counter git checkout
-
Navigate to Application Directory:
cd ./GenAI/AI_PC/Ticket_Counter -
Install the required dependencies:
pip install -r requirements.txt
💡 Make sure you have Miniconda or Anaconda installed before running these commands.
requirements.txt # Lists all Python dependencies needed to run the app
src/
│
├── init_db.py # Initializes the database schema and connections
├── app.py # Main entry point for the Streamlit app
│
└── modules/
│
├── config/
│ └── fares.json # Contains fare configuration data and constants
│
├── pages/
│ ├── home.py # Home page UI and logic
│ ├── workspace.py # Workspace for audio recording and transcription confirmation
│ └── dashboard.py # Dashboard for ticket insights, editing, cancellation, and CSV export
│
└── utils/
├── fare_utils.py # Functions for fare calculation and validation
├── llm_utils.py # Utilities for LLM-based natural language processing
└── transcription_utils.py # Functions for handling audio transcription and error correction
The application provides two main interfaces:
The assistant includes three main pages accessible via sidebar navigation:
🎫 Workspace Page:
- Select Recording duration and input device selection
- Enter the created workspace name
- Enter the API key generated from AnythingLLM
- Transcribe audio using Whisper ONNX.
- Confirm transcription accuracy.
- Extract source, destination, and ticket count using LLaMA.
- View fare, platform, and QR code metadata
- Download ticket metadata
- Access destination travel insights via LLaMA
📊 Dashboard Page:
- View all generated tickets with metadata
- Edit or cancel tickets
- Summarize ticketing activity
- Visualize insights using graphs (e.g., ticket trends, popular routes)
- Download ticket logs and summaries as CSV files
To launch the voice-based ticketing assistant, follow these steps from the root application directory:
This step sets up the necessary tables and schema for ticket storage and fare calculations.
python run src/init_db.pyStart the interactive UI for ticket assistance, voice transcription, and dashboard insights.
streamlit run src/app.pyThis will start the interactive web interface where you can configure and use the application.
- The application uses the Whisper model to transcribe user voice inputs into text.
- This transcribed text is then processed by AnythingLLM models, which generate intelligent responses — including automatic detection and correction of source and destination locations, and provide destination-specific insights.
- To enable this functionality, you must provide a valid API key for the AnythingLLM integration
