FFBack - Financial Feed Backend

CLI to analyze financial news sentiment using finBERT and transformers.js - intended to be used within financialfeeling.com

Principal characteristics

• 200+ news articles from multiple verified RSS sources
• Auto scraping using Cheerio (25 articles with whole content)
• 8 RSS feed configurated by default: NY Times (Business/Tech/Economy), BBC (Business/World), Reddit (WorldNews/Tech/Stocks)
• Intelligent chunking of large articles in 512 char fragments
• Vectorial embeddings using all-MiniLM-L6-v2 for semantic search
• Sentiment analysis with finBERT model specialized in financial sentiment
• Full API REST to integrate easily with Next.js
• Intelligent cache with scraping metadata
• Automatic extraction of from article contents
• Ready for WebLLM! waiting to be implemented in the frontend to further improve the analysis with Qwen AI model (Qwen only working client-side)

Requirements

• Node.js >= 18.x
• npm o yarn
• 4GB+ RAM recommended for ML models

Try the CLI!

./analyze-ticker.sh {TICKER_NAME}

Install

# Clone the repository
git clone <repository-url>
cd ffback

# Install dependencies
npm install

# Cp config file
cp .env.example .env

# Edit config (optional)
nano .env

️Configuración

Edit .env file to customize:

# RSS Feeds - Financial News Sources (VERIFIED)
RSS_FEEDS=https://rss.nytimes.com/services/xml/rss/nyt/Business.xml,https://feeds.bbci.co.uk/news/business/rss.xml,https://feeds.bbci.co.uk/news/world/rss.xml,https://rss.nytimes.com/services/xml/rss/nyt/Technology.xml,https://rss.nytimes.com/services/xml/rss/nyt/Economy.xml,https://www.reddit.com/r/worldnews/.rss,https://www.reddit.com/r/technology/.rss,https://www.reddit.com/r/stocks/.rss

# Scraping Configuration
ENABLE_SCRAPING=true
SCRAPING_DELAY_MS=1200
SCRAPING_MAX_ARTICLES=25

# Cache config (in seconds)
CACHE_TTL=3600

# MAX chunk size for analysis
MAX_CHUNK_SIZE=512
OVERLAP_SIZE=50

# CORS
ALLOWED_ORIGINS=http://localhost:3000,http://localhost:3001

Using it

Development

npm run dev

Production

npm run build
npm start

Other comands

npm run lint    # Ejecutar linter
npm test        # Ejecutar tests

API Endpoints

1. Health Check

GET /api/health

Response:

{
  "status": "ok",
  "timestamp": "2025-10-12T...",
  "uptime": 123.45
}

2. Pipeline status

GET /api/status

Response:

{
  "isProcessing": false,
  "lastUpdateTime": "2025-10-12T...",
  "hasData": true,
  "cacheStats": {
    "keys": 5,
    "hits": 120,
    "misses": 10
  },
  "feeds": [...]
}

3. Analyze Ticker sentiment

POST /api/analyze
Content-Type: application/json

{
  "tickers": ["AAPL", "MSFT", "GOOGL"],
  "timeRange": {
    "start": "2025-10-01T00:00:00Z",
    "end": "2025-10-12T23:59:59Z"
  },
  "sources": ["yahoo", "reuters"]  // opcional
}

Response:

{
  "tickers": [
    {
      "ticker": "AAPL",
      "sentiment": {
        "label": "positive",
        "score": 0.85,
        "confidence": 0.92
      },
      "relatedArticles": [
        {
          "articleId": "abc123",
          "title": "Apple Reports Strong Q4 Results",
          "url": "https://...",
          "publishedAt": "2025-10-11T...",
          "relevanceScore": 0.95
        }
      ],
      "aggregatedSentiment": {
        "overall": "positive",
        "positiveCount": 15,
        "negativeCount": 3,
        "neutralCount": 2,
        "averageScore": 0.78,
        "confidence": 0.88
      },
      "timestamp": "2025-10-12T..."
    }
  ],
  "metadata": {
    "totalArticlesAnalyzed": 150,
    "processingTime": 2345,
    "timestamp": "2025-10-12T..."
  }
}

4. Obtain articles

GET /api/articles?page=1&limit=50

Response:

{
  "articles": [...],
  "pagination": {
    "page": 1,
    "limit": 50,
    "total": 200,
    "totalPages": 4
  }
}

5. Obtain articles by ticker

GET /api/articles/AAPL

Response:

{
  "ticker": "AAPL",
  "articles": [...],
  "count": 25
}

6. Obtain all the tickers

GET /api/tickers

Response:

{
  "tickers": ["AAPL", "MSFT", "GOOGL", ...],
  "count": 150
}

7. Add custom Tickers

POST /api/tickers
Content-Type: application/json

{
  "tickers": ["CUSTOM1", "CUSTOM2"]
}

8. Refesh Pipeline

POST /api/refresh

Force full data update.

Next.Js integration

1. Client API config

// lib/ffback-client.ts
const API_BASE_URL = process.env.NEXT_PUBLIC_API_URL || 'http://localhost:3001/api';

export async function analyzeTickers(tickers: string[]) {
  const response = await fetch(`${API_BASE_URL}/analyze`, {
    method: 'POST',
    headers: {
      'Content-Type': 'application/json',
    },
    body: JSON.stringify({ tickers }),
  });

  if (!response.ok) {
    throw new Error('Failed to analyze tickers');
  }

  return response.json();
}

export async function getArticles(page = 1, limit = 50) {
  const response = await fetch(`${API_BASE_URL}/articles?page=${page}&limit=${limit}`);
  
  if (!response.ok) {
    throw new Error('Failed to fetch articles');
  }

  return response.json();
}

2. Example with React Component

// components/TickerAnalysis.tsx
'use client';

import { useState } from 'react';
import { analyzeTickers } from '@/lib/ffback-client';

export function TickerAnalysis() {
  const [ticker, setTicker] = useState('');
  const [result, setResult] = useState(null);
  const [loading, setLoading] = useState(false);

  const handleAnalyze = async () => {
    setLoading(true);
    try {
      const data = await analyzeTickers([ticker.toUpperCase()]);
      setResult(data);
    } catch (error) {
      console.error('Analysis failed:', error);
    } finally {
      setLoading(false);
    }
  };

  return (
    <div className="p-4">
      <input
        type="text"
        value={ticker}
        onChange={(e) => setTicker(e.target.value)}
        placeholder="Enter ticker (e.g., AAPL)"
        className="border p-2 rounded"
      />
      <button
        onClick={handleAnalyze}
        disabled={loading}
        className="ml-2 bg-blue-500 text-white px-4 py-2 rounded"
      >
        {loading ? 'Analyzing...' : 'Analyze'}
      </button>

      {result && (
        <div className="mt-4">
          {/* Render sentiment results */}
        </div>
      )}
    </div>
  );
}

3. WebLLM integration (Frontend)

Using light models like Qwen3B in the browser:

// lib/webllm-client.ts
import * as webllm from '@mlc-ai/web-llm';

let engine: webllm.MLCEngine | null = null;

export async function initializeWebLLM() {
  if (!engine) {
    engine = await webllm.CreateMLCEngine('Qwen2.5-0.5B-Instruct-q4f16_1-MLC');
  }
  return engine;
}

export async function analyzeWithLLM(tickerData: any, userQuery: string) {
  const engine = await initializeWebLLM();
  
  const prompt = `Given the following sentiment analysis for ${tickerData.ticker}:
${JSON.stringify(tickerData, null, 2)}

${userQuery}`;

  const response = await engine.chat.completions.create({
    messages: [{ role: 'user', content: prompt }],
  });

  return response.choices[0].message.content;
}

️Architecture

┌─────────────────┐
│   RSS Feeds     │
│ (Yahoo, Reuters)│
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  RSS Fetcher    │
│   Service       │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  Normalizer     │
│   Service       │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│   Chunking +    │
│   Embedding     │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│    FinBERT      │
│   Sentiment     │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│   API REST      │
│   + Cache       │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│   Next.js App   │
│   + WebLLM      │
└─────────────────┘

Security aspects

• Helmet.js for security headers (CORS)
• Customizable CORS.
• Entry validation with Zod
• Rate limiting recommended for producción

Performance

• Smart caching with auto expiry
• Processing in chunks to improve efficency
• Cuantized AI models to reduce memory usage
• Periodic updating in the background

Testing

npm test