Ohari — AI Powered Stock Analysis & Prediction System

---

Ohari is a Django-based Stock Market Web Application that combines:

• Real-time stock data
• Stock news integration
• LSTM-based deep learning price prediction
• AI-powered stock assistant
• Admin & User management system
• Complaint & feedback system

This project demonstrates full-stack development with Machine Learning, Sentiment Analysis, and AI integration.

---

Key Features

Authentication & User Roles

• User Registration with profile upload

• Login & Logout system

• Role-based access:

  • Admin
  • User
  • Blocked User

• Change Password

• Forgot Password with OTP (Email-based)

• Profile Editing

---

Admin Functionalities

• View all registered users
• Block / Unblock users
• View and reply to complaints
• View stock-related news
• View real-time stock prices (AAPL, GOOGL, MSFT)
• Change admin password

---

User Functionalities

• View real-time stock prices
• View stock market news (filter by company)
• View historical stock data
• Predict next-day stock price using LSTM
• Send complaints to admin
• View admin replies
• AI Stock Assistant (Ask stock-related questions)
• Edit profile
• Change password

---

Stock Data Integration

Stock prices are fetched using:

• yfinance (Yahoo Finance API)

Tracked stocks (default):

• AAPL
• GOOGL
• MSFT

---

News Integration

News is fetched using:

• NewsAPI
• AlphaVantage (for sentiment-based news)

Users can filter news by companies:

• Tesla
• Apple
• Microsoft
• Google
• Amazon
• Nvidia
• Facebook
• IBM
• Infosys

---

LSTM Stock Price Prediction

The prediction pipeline includes:

1️ Historical Data Fetching

• 6 months stock data using yfinance

2️ Sentiment Analysis

• News sentiment extracted using TextBlob
• Polarity score added as feature

3️ Data Preparation

• MinMax Scaling

• Lookback window = 60 days

• Features:

  • Close Price
  • Sentiment Score

4️ Model Architecture

LSTM (50 units, return_sequences=True)
Dropout (0.2)
LSTM (50 units)
Dropout (0.2)
Dense (25)
Dense (1)

Optimizer: Adam Loss: Mean Squared Error Epochs: 20

5️ Output

• Predicts next day closing price
• Generates matplotlib graph
• Graph converted to Base64 and embedded in HTML

---

AI Stock Assistant

Integrated with:

• OpenRouter API
• DeepSeek Model (deepseek-r1-distill-qwen-32b)

Users can:

• Ask stock-related queries
• Get AI-generated financial insights

Response is returned as formatted HTML via AJAX.

---

Project Modules Overview

Authentication

• Login
• Register
• Forgot Password (OTP via Email)
• Session Management

Admin Panel

• User Management
• Complaint Management
• News Viewing

User Dashboard

• Stock Prices
• Historical Data
• Prediction Engine
• AI Assistant
• Profile Management

---

Technology Stack

Backend

• Django
• Python

Machine Learning

• TensorFlow / Keras
• LSTM Networks
• Scikit-learn (MinMaxScaler)
• TextBlob (Sentiment Analysis)

Data APIs

• Yahoo Finance (yfinance)
• NewsAPI
• AlphaVantage

Frontend

• HTML
• CSS
• Django Templates
• JavaScript (AJAX for AI assistant)

Visualization

• Matplotlib

Database

• Django ORM (SQLite by default)

---

Installation Guide

1️ Clone the Repository

git clone https://github.com/ThomasMathewz/Ohari.git
cd Ohari

2️ Create Virtual Environment

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

3️ Install Requirements

pip install -r requirements.txt

---

Setup Environment Variables

You must configure:

• NewsAPI Key
• AlphaVantage API Key
• OpenRouter API Key
• Email settings (for OTP system)

Update in:

• settings.py
• Views where API keys are used

Important: Never expose API keys in production.

---

Run the Project

python manage.py makemigrations
python manage.py migrate
python manage.py runserver

Visit:

http://127.0.0.1:8000/

---

Prediction Flow Summary

User selects stock → Fetch historical data → Fetch sentiment → Prepare dataset → Train LSTM → Predict next-day price → Display result + Graph

---

Security Notes

• Passwords are stored as plain text (Needs improvement)
• API keys are hardcoded (Should be moved to environment variables)
• Model retrains on every prediction (Can be optimized)

This project is primarily for academic and demonstration purposes.

---

Project Domain

• Artificial Intelligence
• Deep Learning
• FinTech
• Web Development
• Time Series Forecasting
• NLP (Sentiment Analysis)

---

Future Improvements

• Use Django authentication system with password hashing
• Store trained LSTM model instead of retraining each request
• Add stock portfolio management
• Improve UI/UX
• Deploy on cloud (AWS / Render / Railway)
• Add REST APIs

---

Author

Thomas Mathew

B.Tech Graduate

Aspiring AI & Software Developer

---