📊 EquiSense — AI-Powered Market Sentiment Analysis

Real-time financial sentiment intelligence for retail investors — democratizing the market analysis tools that institutions pay lakhs for.

---

🎯 Problem Statement

90 million retail investors in India make investment decisions based on incomplete information.

While institutional traders have access to Bloomberg terminals, quant teams, and real-time sentiment feeds costing ₹10-15 lakhs annually, retail investors rely on fragmented news sources, social media noise, and guesswork.

EquiSense bridges this gap — providing institutional-grade sentiment analysis powered by FinBERT NLP and Gemini AI, accessible to anyone with an internet connection.

---

✨ Key Features

🧠 Financial NLP with FinBERT

• Sentiment classification trained specifically on financial language
• Understands context (e.g., "rate cut" is bullish, not bearish)
• 50-article batch processing in seconds

📰 Multi-Source News Aggregation

• GNews, NewsAPI, Finnhub, Bloomberg RSS, CNBC RSS
• Real-time data from 5+ sources in parallel
• Automatic deduplication and recency weighting

📊 Visual-First Dashboard

• Sentiment Gauge: Interactive Fear ↔ Greed indicator (-100 to +100)
• Donut Chart: Fear/Neutral/Greed distribution breakdown
• Source Analysis: Horizontal bar chart with article counts
• Signal Cards: Color-coded bullish/bearish/neutral alerts
• SRI Progress Bar: Sentiment Reliability Index (0-100)

🎯 Advanced Analytics

Sentiment Reliability Index (SRI)

Weighted composite score combining:

• Source credibility (Bloomberg 1.0 → Random aggregator 0.6)
• Article recency (6hr decay curve)
• Model confidence (FinBERT score)

Market Regime Detection

Classifies current environment using VIX + sentiment dispersion:

• Risk-On: Low volatility + consensus sentiment
• Risk-Off: High VIX + high dispersion
• Event-Driven: Mixed signals

Smart Money Divergence Alerts

Flags when sentiment and price action decouple:

• Bullish news + falling prices → Distribution risk
• Bearish news + rising prices → Capitulation signal

🤖 AI Analysis with Gemini 2.5 Flash

• Structured analyst-style summaries in 250 words
• Sentiment classification (Strongly Bullish → Strongly Bearish)
• Key insights + suggested action (Accumulate/Hold/Book Profits/Avoid)

---

🚀 Live Demo

Try it here: equisense.streamlit.app

🛠️ Tech Stack

Component	Technology	Purpose
Frontend	Streamlit	Interactive web dashboard
NLP Model	FinBERT (ProsusAI)	Financial sentiment classification
AI Analysis	Gemini 2.5 Flash	Natural language synthesis
Visualization	Plotly	Interactive charts (gauge, donut, bars)
Market Data	yfinance	NIFTY 50, India VIX, real-time prices
News APIs	GNews, NewsAPI, Finnhub	Multi-source news aggregation
RSS Feeds	feedparser	Bloomberg, CNBC financial news

---

📖 How It Works

Workflow

1. Enter a search term (e.g., "NIFTY 50", "Reliance Industries", "IT Sector")
2. Multi-source data aggregation: Fetches 30-50 articles from 5 news sources in parallel
3. FinBERT sentiment classification: Analyzes each headline for positive/negative/neutral sentiment
4. Advanced analytics: Calculates SRI, detects market regime, flags divergences
5. AI synthesis: Gemini 2.5 Flash generates a structured analyst summary
6. Visual dashboard: Presents insights through interactive charts and color-coded signals

Reading the Dashboard

Overview Tab

• Sentiment Gauge: Net sentiment from -100 (extreme fear) to +100 (extreme greed)
• Donut Chart: Percentage breakdown (Fear / Neutral / Greed)
• Signal Card: Current market signal with color coding
• Key Metrics: NIFTY 50 price, India VIX, Put-Call Ratio
• SRI Bar: Confidence level (High/Medium/Low)
• AI Analysis: Gemini-generated summary

Analysis Tab

• Source Breakdown: Which news outlets were analyzed
• Top Contributors: Most influential positive/negative headlines
• Directional Alignment: Does sentiment match price action?

Data Tab

• Articles: Full list of analyzed headlines with links
• Sentiment Data: Raw FinBERT classification results
• Limitations: Known issues and disclaimers

Signal Interpretation

Signal	Meaning	Implication
🔥 Extreme Panic	Fear > 70%, VIX > 20	Potential reversal zone
📉 High Fear	Fear 55-70%	Cautious buying opportunity
⚖️ Neutral Zone	Balanced sentiment	Hold positions
📈 High Greed	Greed 55-70%	Consider profit booking
⚠️ Extreme Euphoria	Greed > 70%	Distribution risk

---

🧪 Architecture

┌─────────────────────────────────────────────────────────────┐
│                     User Interface (Streamlit)              │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐  │
│  │ Overview │  │ Analysis │  │   Data   │  │ Sidebar  │  │
│  └──────────┘  └──────────┘  └──────────┘  └──────────┘  │
└───────────────────────────┬─────────────────────────────────┘
                            │
┌───────────────────────────▼─────────────────────────────────┐
│                     Data Aggregation Layer                   │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐  ┌─────────┐       │
│  │  GNews  │  │ NewsAPI │  │ Finnhub │  │   RSS   │       │
│  └─────────┘  └─────────┘  └─────────┘  └─────────┘       │
└───────────────────────────┬─────────────────────────────────┘
                            │ (50 articles)
┌───────────────────────────▼─────────────────────────────────┐
│                    Sentiment Analysis Engine                 │
│  ┌──────────────────────────────────────────────────────┐  │
│  │  FinBERT (ProsusAI/finbert)                          │  │
│  │  - Batch classification (positive/negative/neutral)  │  │
│  │  - Confidence scores (0-1)                           │  │
│  │  - Keyword fallback (if model unavailable)           │  │
│  └──────────────────────────────────────────────────────┘  │
└───────────────────────────┬─────────────────────────────────┘
                            │
┌───────────────────────────▼─────────────────────────────────┐
│                   Advanced Analytics Layer                   │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐        │
│  │     SRI     │  │   Regime    │  │ Divergence  │        │
│  │  Scoring    │  │  Detection  │  │    Alert    │        │
│  └─────────────┘  └─────────────┘  └─────────────┘        │
└───────────────────────────┬─────────────────────────────────┘
                            │
┌───────────────────────────▼─────────────────────────────────┐
│                  AI Synthesis (Gemini 2.5 Flash)             │
│  - Structured prompt (250 words)                            │
│  - Sentiment summary + Key insights + Outlook + Action      │
└─────────────────────────────────────────────────────────────┘

Key Algorithms

Sentiment Reliability Index (SRI)

SRI = (source_credibility × 0.4) + (recency_weight × 0.3) + (model_confidence × 0.3)

where:
  source_credibility ∈ [0.6, 1.0]  (Bloomberg = 1.0, GNews = 0.65)
  recency_weight     ∈ [0.3, 1.0]  (6h = 1.0, 72h = 0.5)
  model_confidence   ∈ [0.0, 1.0]  (FinBERT output score)

Market Regime Classification

if VIX > 18 and dispersion > 0.35:
    return "Risk-Off"
elif dispersion > 0.40 and 12 < VIX <= 18:
    return "Event-Driven"
elif VIX < 12 and dispersion < 0.25:
    return "Risk-On"
else:
    return "Event-Driven"

---

📊 Performance Benchmarks

Metric	Value	Notes
Analysis Time	10-15s	50 articles, full pipeline
API Calls	5-7	Parallel execution
FinBERT Accuracy	~85%	On financial headlines
SRI Calibration	±8 points	Across different market regimes
Cache Hit Rate	~60%	15min TTL on news data

---

🔮 Future Roadmap

Potential enhancements being considered:

• Portfolio-level sentiment tracking
• WhatsApp/Telegram alert integration
• Real-time options flow data (NSE)
• Historical backtesting with real data
• Mobile app (React Native / Flutter)
• Multi-language support (Hindi, regional languages)

---

⚠️ Known Limitations

Model Limitations

• FinBERT may misclassify sarcasm or domain-specific jargon
• Headline-only analysis misses context from full articles
• Keyword fallback has significantly lower accuracy (~60%)

Data Limitations

• RSS feeds can be stale or contain duplicates
• API rate limits reduce sample size during high-traffic periods
• Source credibility weights are heuristic, not empirically validated

Market Indicator Limitations

• PCR is simulated — real-time NSE data not available via free APIs
• NIFTY/VIX data depend on exchange hours and yfinance availability
• Sentiment-price correlation breaks during regime changes

Backtesting Caveats

• Simulated accuracy is illustrative, not predictive
• No transaction costs, slippage, or market impact modeled
• Limited lookback period (does not capture black swan events)

---

📜 Disclaimer

This tool is for educational and research purposes only.

EquiSense is NOT:

• Financial advice
• An investment recommendation
• A trading signal
• A substitute for professional financial consultation

Important Notes:

• Past sentiment patterns do not guarantee future performance
• Market conditions can change rapidly, invalidating prior analysis
• Always consult a registered financial advisor before making investment decisions
• The developers assume no liability for financial losses incurred

---

🙏 Acknowledgments

• ProsusAI — FinBERT model
• Google — Gemini 2.5 Flash API
• Streamlit — Rapid prototyping framework
• Plotly — Interactive visualizations
• yfinance — Market data access
• HuggingFace — Transformers library

---

📧 Contact

For inquiries or collaboration opportunities:

• Email: harshprabha95@gmail.com
• LinkedIn: MY Profile
• Portfolio: My Website

---

Built with ❤️ using Python, Streamlit, and AI

© 2026 [Harshprabha]. All rights reserved.