QuickStock™ is a state-of-the-art inventory management and billing system designed specifically for the modern Indian retail landscape. By leveraging Voice Intelligence (STT) and Natural Language Understanding (NLU), it allows shopkeepers to manage stock and process bills using natural speech in English, Hindi, and Hinglish.
This project bridges the gap between traditional manual record-keeping and high-tech retail automation, providing a seamless, "hands-free" experience that reduces transaction time and minimizes human error.
- 🎙️ Real-time Bilingual Voice Commands: Manage inventory using voice in Hindi, English, or a mix of both (Hinglish). Powered by Vosk ASR.
- 📊 Interactive Analytics Dashboard: Visualize sales trends, stock levels, and revenue metrics with high-performance charts (Chart.js).
- 🧠 Smart NLU Engine: Automatically extracts products, quantities, and units from natural speech using spaCy NER and custom Regex patterns.
- 📱 Omnichannel Notifications:
- WhatsApp Integration: Automated restock alerts and bill confirmations via Facebook Graph API.
- SMS OTP: Secure login using Fast2SMS for mobile-first authentication.
- 🛡️ Enterprise-Grade Security: JWT-based authentication, rate limiting, and encrypted session management.
- 📍 Location-Based Intelligence: Find nearby wholesale partners and shops using PostGIS spatial queries.
- 📑 Automated Summarization: Generate concise daily transaction summaries using the LexRank algorithm.
- Core: Python 3.10+ / Flask
- Real-time: Flask-SocketIO (Eventlet)
- Database: PostgreSQL with PostGIS extensions
- Security: Flask-JWT-Extended, Flask-Limiter, Cryptography
- Speech-to-Text: Vosk (Offline HMM+DNN models)
- Natural Language: spaCy (NER), NLTK, Indic NLP Library
- Translation: Deep Translator (Google & MyMemory)
- Matching: Difflib (Ratcliff-Obershelp fuzzy matching)
- Summarization: Sumy (LexRank summarizer)
- Design: Vanilla CSS3 (Modern Glassmorphism Aesthetic)
- Logic: ES6+ JavaScript
- Visuals: Chart.js, Remix Icons, Google Fonts (Outfit)
QuickStock™ implements several advanced algorithms to ensure high accuracy and performance:
- Vosk STT (Kaldi-based): Uses Hidden Markov Models (HMM) and Deep Neural Networks (DNN) for offline speech recognition.
- Hinglish Merging Logic: A confidence-based selection algorithm that merges Hindi and English tokens from dual-recognizer streams.
- Hindi Number Word Parsing: A recursive descent parser that converts phrases like "दो सौ पचास" to
250in real-time. - LexRank Summarization: A graph-based ranking algorithm (similar to Google's PageRank) to extract key transaction insights.
- ASR Error Correction: Persistent N-gram fuzzy matching to correct common speech recognition errors against the product database.
├── app.py # Main Flask application & NLU logic
├── voice_stream_file.py # Real-time Voice/STT processing client
├── static/ # CSS, JS, and UI assets
├── templates/ # Responsive HTML5 templates
├── tools/ # Utility scripts for DB and migration
├── models/ # Vosk language models (Hi/En-IN)
├── ALGORITHMS_USED.md # Detailed technical documentation
└── LIBRARIES_USED.md # Dependency breakdown
- Python 3.10+
- Docker & Docker Compose
- Vosk Models (
vosk-model-en-in-0.5,vosk-model-hi-0.22)
QuickStock uses PostGIS for location-based analytics. You can quickly spin up a database instance using the included docker-compose.yml:
docker-compose up -dCreate a .env file and add your database URL:
DATABASE_URL=postgresql://postgres:postgres_password@localhost:5432/quickstock-
Install dependencies:
pip install -r requirements.txt python -m spacy download en_core_web_sm
-
Initialize Database Schema:
python setup_db.py python migrate_to_db.py
-
Run the application:
python app.py
Developed by Glavin | © 2026 QuickStock™