🛒 LocalLens: Hyper-Local Inventory Intelligence Engine

LocalLens is an end-to-end data science application designed to bridge the gap between local grocers and modern consumers. It serves as a dual-sided platform: a Smart Recipe Finder for consumers and a AI-Powered Operating System for shopkeepers.

---

📸 Screenshots

(Add screenshots here later! Drag and drop your images of the Dashboard, Deep Dive, and POS pages here)

Triage Dashboard	Deep Dive Analytics

---

🚀 Key Features

🏢 For The Business (The Shopkeeper OS)

• 🧠 Automated Demand Triage: Runs 150+ time-series models (Prophet) daily to identify items at risk of stockouts.
• 📈 Deep Dive Analytics: Visualizes daily sales forecasts, identifying trends vs. seasonality.
• 💰 Digital Point-of-Sale (POS): A functional cash register interface that records real-time sales and updates inventory instantly.
• 📦 Smart Inventory Management: Purchase Order (PO) generator that calculates safety stock and suggests reorder quantities automatically.

🛒 For The Consumer

• 🍳 NLP Recipe Parser: Paste any URL from Food.com or Allrecipes. A custom-trained spaCy NER model extracts ingredients (ignoring quantities/prep instructions) and finds them in local stock.
• 🔍 Fuzzy Search Engine: Uses SQL fuzzy matching to link complex recipe ingredients (e.g., "finely chopped organic spinach") to store inventory ("Frozen Spinach").

---

🛠️ The Tech Stack

• Database: PostgreSQL (Relational Data Warehousing)
• Backend & App Logic: Python, Pandas, NumPy
• Machine Learning (Forecasting):
  • Library: Facebook Prophet
  • Features: Seasonality, Indian Holidays, External Regressors (Google Trends data), Promotional events ("On Sale" flags).
• Natural Language Processing (NLP):
  • Library: spaCy (Custom trained NER model)
  • Task: Named Entity Recognition for ingredient extraction.
• Frontend: Streamlit (Multi-page application framework)
• Visualization: Plotly & Altair (Interactive charts)

---

⚙️ How to Run Locally

Prerequisites

• Python 3.10+
• PostgreSQL installed and running locally.

Installation

1. Clone the Repository

   git clone [https://github.com/sdiv0503/Local-Lens.git](https://github.com/sdiv0503/Local-Lens.git)
   cd Local-Lens

2. Set up Virtual Environment

   python -m venv venv
   # Windows:
   .\venv\Scripts\activate
   # Mac/Linux:
   source venv/bin/activate

3. Install Dependencies

   pip install -r requirements.txt

4. Database Setup

   • Create a PostgreSQL database named local_lens_db.
   • Update db.py with your database password.
   • Run the setup scripts to seed the DB with synthetic data:

   # 1. Create Tables
   # (Manually run content of schema.sql in your SQL tool)
   
   # 2. Populate Stores & Products
   python populate_db.py
   
   # 3. Generate Fake Trends & Simulation Data
   python create_fake_trends.py
   python simulate_sales.py
   
   # 4. Train the ML Models
   python train_all_models.py

5. Run the App

   streamlit run app.py

---

📊 Data Science Methodology

1. Synthetic Data Generation

Since real-world retail data is private, I built a robust Data Simulator (simulate_sales.py) that generates 5 years of daily sales history.

• Logic: Sales = Base_Demand + Seasonality (Sine Wave) + Trend_Influence + Promotions + Noise.
• This ensures the ML models have realistic patterns (like Mango spikes in summer) to learn from.

2. Custom NLP Training

Standard NLP models fail on recipe text (e.g., confusing "can" the noun vs "can" the verb).

• Solution: Manually labeled 100+ recipe strings.
• Outcome: A custom spaCy model that achieves high accuracy in extracting the core buyable ingredient from complex text.

3. Forecasting at Scale

Instead of one model, the system trains 150+ individual models (one per product).

• Optimization: Models are trained in batch (train_all_models.py) and serialized to JSON.
• Inference: The dashboard loads these lightweight JSONs for instant, on-demand forecasting.

---

📜 License

Distributed under the MIT License. See LICENSE for more information.

---

Built with ❤️ by Divyansh Sharma