🌐 Financial Intelligence Hub – Backend

A centralized platform for smarter, data-driven financial decisions.

This project is a Backend API that powers a financial analytics platform designed to centralize market information and help users make informed investment decisions.
It exposes structured endpoints that aggregate financial news through web scraping, process market data for stocks, ETFs, and currencies, compute technical indicators, recommend daily trading opportunities, and provide a multilingual AI chatbot capable of answering questions about both the platform and the financial world.

---

🚀 Key Features

• Financial News Aggregation from multiple sources using web scraping.
• Market Charts & Indicators for stocks, ETFs, and forex — all market data sourced from Yahoo Finance (price history, RSI, MA, volatility, etc.).
• Trade of the Day Engine recommending the strongest opportunities of the day based on RSI-driven analysis, enriched with insights from a third-party financial analysis provider.
• AI Chatbot (EN/ES) trained with NLP/ML models to answer financial and platform-related questions.
• Centralized API with MVC Architecture serving all data to the React frontend.

---

🔗 Live Demo & Frontend Repo

• 🌐 Live Web App: https://d3iqcabelr1yf0.cloudfront.net/app/market
• 👤 Demo User: [email protected]
• 🔑 Password: demo123456
• 📦 Frontend Repository:
  https://github.com/SebastianGranadosJ/financial-intelligence-hub-front

---

Infrastructure Diagram

📰 Financial News Module

This module centralizes financial news from multiple sources to give users a concise, sentiment-aware overview of the market. Each news entry includes:

• Title
• Description
• Publication date
• Source
• Original link
• Sentiment score (positive/negative + intensity)

🔍 How It Works

1. Targeted News Discovery

The system builds a pool of news URLs using Google News queries generated from:

• A predefined list of financial topics
• A predefined set of seed sources

Each query produces a Google News results page, which becomes the input for scraping.

2. Web Scraping Pipeline

For every Google News results page, the scraper extracts key elements using Google’s specific selectors:

Element	CSS Selector
Title	div.n0jPhd
URL	a.WlydOe
Description	div.GI74Re
Date	div.rbYSKb
Source	div.NUnG9d

The scraper normalizes the extracted data and prepares it for NLP processing.

3. Sentiment Analysis Using DistilBERT

Each article description is analyzed using a DistilBERT model from Hugging Face Transformers.

Why DistilBERT?

• ~97% of BERT’s accuracy
• ~40% fewer parameters
• Significantly faster inference (ideal for large scraping batches)

The model returns a sentiment label (positive/negative) and a confidence score.

4. Smart Storage & Duplicate Prevention

All processed news items are stored in MongoDB along with their sentiment and scrape metadata.
Before saving, the system filters out articles already scraped or older than the last session, ensuring efficient crawling with no duplicates.

📈 Financial Market Module

This module provides price charts, metrics, and technical indicators for Stocks, ETFs, and Forex, enabling users to analyze trends and make informed decisions.

🔹 Asset Pools

Stocks:

• S&P 500 constituents

ETFs: 15 key ETFs representing indices, sectors, commodities, and strategies:

• SPY – S&P 500 ETF
• VOO – Vanguard S&P 500 ETF
• QQQ – Nasdaq 100 ETF
• DIA – Dow Jones Industrial Average ETF
• IWM – Russell 2000 ETF
• EFA – MSCI EAFE ETF
• EEM – Emerging Markets ETF
• XLF – Financials Select Sector ETF
• XLK – Technology Select Sector ETF
• XLE – Energy Select Sector ETF
• XLV – Health Care Select Sector ETF
• GLD – Gold ETF
• SLV – Silver ETF
• ARKK – ARK Innovation ETF
• VTI – Vanguard Total Stock Market ETF

Forex: 10 highly liquid major and popular cross pairs covering ~95% of global volume:

• Majors: EUR/USD, USD/JPY, GBP/USD, USD/CHF, AUD/USD, USD/CAD, NZD/USD
• Crosses: EUR/GBP, EUR/JPY, GBP/JPY

---

🔹 Data Storage & Updates

• Metrics: fetched from Yahoo Finance and updated several times daily (no historical storage).
• Time Series: stored historically at daily granularity in PostgreSQL; last 5 years of data preloaded via script.

Query Behavior:

• 1 Month / 1 Year / 5 Years: data retrieved from PostgreSQL
• 1 Day / 5 Days: data fetched directly from Yahoo Finance via a Facade pattern for clean abstraction and API consistency

---

🔹 Features

• Visualize historical price charts for all assets.
• Access key metrics and technical indicators for informed decision-making.
• Optimized for fast updates on short-term queries while preserving historical context for long-term analysis.

🤖 AI Chatbot Module

This module provides a bilingual (EN/ES) chatbot that answers frequently asked questions about the platform and basic financial concepts.

🔹 Scope

• Platform FAQs: 18 predefined questions, e.g.:

  • What is StarkAdvisor?
  • What limits does StarkAdvisor have?
  • ¿Quién hizo esto? / Who made StarkAdvisor?
  • Hello Starky
  • ¿Qué significa Trade del Día?

• Financial Concepts: 24 predefined questions, e.g.:

  • What is a stock?
  • What is a bond?
  • What is an ETF?
  • What is market capitalization?
  • What is a commodity?
  • What is ROI?

Each question type is treated as a class, and the bot is trained to handle multiple variations of each question.

---

🔹 How It Works

1. Question Normalization:

   • Input is lemmatized using spaCy to standardize wording.
   • The spaCy model is loaded via Hugging Face Transformers.

2. Language Detection:

   • The bot detects whether the question is in English or Spanish.

3. Classification & Response:

   • The normalized question is passed to the classifier, which determines the question type.
   • If the bot is confident, it returns the predetermined answer.
   • If uncertain, the failover module provides a default response.

4. Logging:

   • All received questions and responses are stored in PostgreSQL for monitoring and analytics.

---

🔹 Training

• The chatbot was trained with multiple variations per question to improve classification robustness.
• Full training resources and scripts are available at:
  Mbert Chatbot Training Repository

---

💹 Trade of the Day Module

This module provides an automated daily investment recommendation system that highlights the best trades of the day based on quantitative analysis.

🔹 How It Works

• The system retrieves the top 10 trades of the day from a Hugging Face API.

• Each trade includes the following parameters:

  • Ticker: asset symbol
  • Expected Return (%): projected gain over the next 21 days
  • RSI (Relative Strength Index): momentum indicator
  • Description: explanation of why it is considered a good investment

• These recommendations are updated daily and presented via the backend API for the frontend to display.

---

🔹 Features

• Automated, data-driven trade suggestions.

• Combines technical indicators and external insights for each recommendation.

• Provides concise reasoning for each suggested trade to help users understand the rationale.

• API reference for the daily trades: Hugging Face Trade API