Skip to content

About

AI-driven equity intelligence platform featuring Brownian Bridge chart dynamics, FinBERT sentiment analysis, ML-powered BUY/SELL signals, and real-time market insights — built with Flutter & FastAPI.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

Header

Status NSE Universe Accuracy Speed Paper Capital Release

Flutter Dart Riverpod FastAPI Python XGBoost Groq AI Syncfusion License: MIT


TradeVision AI is an institutional-grade, cross-platform stock market analytics and trading intelligence ecosystem engineered specifically for the Indian Equities Market (NSE & BSE). It bridges the gap between raw market noise and actionable conviction by unifying Universal XGBoost v2 machine learning models, an Explainable AI (XAI) reasoning waterfall, Groq Llama 3 70B financial copiloting, FinBERT news sentiment scoring, an anchored Brownian Bridge stochastic charting engine, and an interactive Virtual Paper Trading Simulator inside a high-performance Flutter client.


Explore Codebase API Docs Viva Guide Research Paper

  ┌────────────────────────────────────────────────────────────────────────────────────────┐
  │  🚀 HIGH-CONVICTION ML SIGNALS  │  ⚡ SUB-SECOND GROQ COPILOT  │  📈 LIVE NSE CANDLESTICKS │
  └────────────────────────────────────────────────────────────────────────────────────────┘

✨ Key Features • 🏛️ System Architecture • 🧠 Machine Learning • 🤖 AI Copilot • 📈 Stochastic Charts • 💼 Paper Trading • 🎓 Viva & Research • 🛠️ Built With • 🚀 Quick Start • 📡 API Reference • 👥 Contributors


📊 Live Metrics & Platform Scorecard

Metric Benchmark Score Target Threshold Production Status
Gated Directional Precision 97.19% > 90.00% 🟢 Verified on Out-of-Sample NSE Data
Area Under ROC Curve (ROC-AUC) 0.9869 > 0.9500 🟢 Calibrated Dual-Sigmoid Thresholds
XGBoost Inference Latency < 12 ms < 50 ms ⚡ Ultra-Fast Local Vector Computation
Groq LPU Generation Speed 300+ T/s > 150 T/s 🚀 Sub-Second Token Stream
Bundled Equity Universe 2,500+ NIFTY 500+ 🏢 Zero-Latency Offline Indexed Equities
Simulated Capital Sandbox ₹10,00,000 Real Margin 💼 Persistent Virtual Portfolio Ledger
Final Android Package (APK) v1.0.0+2 Universal Fat APK 📱 Core Library Desugaring & Java 17

⚔️ Why TradeVision AI?

Architectural Dimension Traditional Trading Apps (Groww / Zerodha) ⚡ TradeVision AI Experience
Market Intelligence ⚠️ Raw financial headlines without risk context ✅ XGBoost v2 ML Engine + Groq LPU synthesis with XAI feature weights
Chart Mathematical Accuracy ❌ Arbitrary random walks or synthetic drift ✅ Syncfusion Candlesticks + Brownian Bridge stochastic anchoring at 60 FPS
Predictive Transparency ❌ Opaque recommendations or black boxes ✅ XAI Feature Waterfall revealing RSI, MACD, Volume & EMA contributions
Adaptive Financial Assistant ❌ Scripted FAQ bots or zero conversational memory ✅ Groq Copilot with 3 Scaling Modes (Quick, Standard, Deep)
Risk-Free Practical Sandbox ⚠️ Real money required to understand volatility ✅ Virtual Paper Trading Simulator with ₹10,00,000 in simulated capital
Institutional Privacy ❌ Sensitive balances visible in app switcher ✅ App Switcher Privacy Shield with automatic canvas blur
Trading Session Engine ❌ Static device clock ✅ Live Indian Standard Time (IST) Engine with Pre-Open & Closing phases

✨ Key Features

  ┌────────────────────────────────────────────────────────────────────────────────────────┐
  │                                PLATFORM CAPABILITY MATRIX                              │
  ├────────────────────────────────────────────────────────────────────────────────────────┤
  │  [01] Universal XGBoost v2 Engine    │ 27 features, dual-sigmoid gating, XAI waterfall │
  │  [02] Groq Llama 3 70B Copilot       │ Quick (250t), Standard (650t), Deep (1500t)     │
  │  [03] Syncfusion Pro Candlesticks    │ Interactive pinch-to-zoom, crosshairs, overlays │
  │  [04] Brownian Bridge Chart Engine   │ Anchored continuous Wiener process at 60 FPS    │
  │  [05] Indian Market Session Sync     │ 09:15–15:30 IST uniform 1-hr axis across screens│
  │  [06] Production Notification Hub    │ Rate-limited, deduplicated, zero-storm startup  │
  │  [07] Virtual Paper Trading Sandbox  │ ₹10,00,000 capital, instant execution, live P&L  │
  │  [08] FinBERT Real-Time Sentiment    │ Scraped Indian news headlines, NLP scoring      │
  │  [09] Institutional Security Shield  │ App switcher blur, encrypted vault & native icon│
  │  [10] Live IST Market Clock          │ Auto Pre-Open, Regular, Post-Market detection   │
  │  [11] OLED Dark & Clean Light Themes │ 60 FPS 3D Matrix4 card flip animations & haptics│
  └────────────────────────────────────────────────────────────────────────────────────────┘

🧠 1. Universal XGBoost v2 Market Intelligence

  • 27 Engineered Technical Features: Evaluates multi-timeframe momentum, trend velocity, volatility metrics, volume ratios, OBV slopes, VWAP deviations, and session phase dynamics.
  • Calibrated Gating: Signals are classified as BUY, SELL, or HOLD with mathematical probability estimates and confidence levels (HIGH, MEDIUM, LOW).
  • Explainable AI (XAI): Visual waterfall breakdown illustrating the exact contribution of each technical indicator to the final conviction score.

🤖 2. Groq Llama 3 70B Copilot with 3 Depth Scaling Modes

  • ⚡ Quick Mode (250 tokens): Ultra-fast executive summary covering prevailing sentiment, key support/resistance levels, and immediate directional bias.
  • 📊 Standard Mode (650 tokens): Balanced multi-factor breakdown evaluating moving averages, RSI divergence, MACD momentum, and volume dynamics.
  • 🔬 Deep Mode (1500 tokens): Institutional-grade comprehensive brief with risk-reward ratios, stop-loss calculations, multi-target milestones, and bull/bear scenario planning.
  • 🧭 Conversational Routing & App Guidance: Recognizes user greetings, extracts names, and provides complete step-by-step navigation guides for all in-app tabs and analytical tools.

📈 3. Anchored Brownian Bridge Chart Engine & Indian Market Hours Sync

  • Indian Market Session Synchronization (09:15 to 15:30 IST): 1D intraday charts are anchored strictly to official NSE/BSE market trading hours, starting at 09:15 AM IST with uniform 1-hour interval ticks (09:15, 10:15, 11:15, 12:15, 13:15, 14:15, 15:15) across all screens (Stock Detail, Market & Charts, Quick Chart, Analysis).
  • Live OHLC HUD & Crosshairs: Real-time trackball and crosshair snapping with continuous OHLC display in Indian Rupee format (₹#,##0.00).
  • Elimination of Synthetic Drift: Interpolates discrete server-side OHLC bars into fluid, realistic micro-movements pinned precisely to exchange open and live market close: $$B(t) = W(t) - \frac{t}{T} W(T)$$
  • Dual-Harmonic Correction: Injects intraday trading wave rhythms (morning discovery, midday consolidation, and closing run) without violating price boundaries.
  • Syncfusion Interactive Candlesticks: High-performance rendering with smooth pinch-to-zoom, crosshair snapping, volume distribution bars, SMA(20), EMA(50), Bollinger Bands (20, 2), RSI(14), and MACD(12, 26, 9).

⚡ 4. Production-Grade Notification & Milestone Hub

  • Real-Time Market Milestone Precision: Monitored via IST clock for accurate session events: Pre-Open (09:00 AM), Market Open (09:15 AM), Regular Market Close (03:30 PM), and Post-Market Settlement (04:00 PM IST).
  • Zero-Storm Startup Suppression: On application launch, initial news headlines and pre-existing paper portfolio positions are silently indexed within a 15-second grace window, completely eliminating startup notification floods.
  • Deduplication & Rate Limiting: Persistent StorageService milestone tracking prevents duplicate alerts across app restarts within the same calendar day, backed by an in-memory 80-item deduplication buffer and a 4-second burst throttle.

💼 5. Virtual Paper Trading Simulator

  • Instant Virtual Capital: Trade Indian equities with ₹10,00,000 in simulated margin.
  • Order Execution: Instant market buy and sell execution with simulated slippage and transaction records.
  • Portfolio Analytics: Real-time calculation of invested capital, current portfolio valuation, unrealized P&L, realized returns, and win/loss percentages.
  • Local Persistence: Secure local caching of transactions and open positions across application launches.

🛡️ 6. Institutional Security & Privacy

  • App Switcher Privacy Shield: Minimizing the app or opening the Android task switcher automatically renders a privacy blur to shield portfolio balances and orders.
  • Dedicated Native Notification Icon: High-importance Android notification channel with a dedicated stencil icon (ic_notification.png) for live price breakout alerts.
  • Encrypted Vault Storage: Token masking and local encryption protecting offline paper trading states.

📰 7. Real-Time News & FinBERT Sentiment Analysis

  • Live News Aggregation: Scrapes and streams curated financial headlines for top Indian conglomerates.
  • FinBERT Sentiment Scoring: Financial-domain NLP model classifies incoming headlines into Bullish, Bearish, or Neutral with confidence percentages.
  • Conflict Override Engine: If technical momentum indicates bullish continuation but breaking news sentiment scores sharply bearish, the platform safely flags an emergency HOLD.

🛠️ Built With

My Skills


Layer Technologies & Libraries Version / Specs
Mobile & Web Client Flutter, Dart, Riverpod 2.x, GoRouter, Google Fonts (Inter, Roboto Mono) Flutter 3.x / Dart 3.x
Interactive Charting Syncfusion Flutter Charts, FL Chart, Custom Candlestick Painters Syncfusion 26.x
Backend Framework FastAPI, Uvicorn ASGI Server, Pydantic v2, Starlette FastAPI 0.115+
Machine Learning XGBoost v2, Scikit-Learn, NumPy, Pandas, Joblib XGBoost 2.0+
AI Copilot Inference Groq Cloud LPU, Meta Llama-3.3-70b-versatile Engine 300+ Tokens/Sec
Financial NLP FinBERT, Transformers, HuggingFace Hub NLP Polarity Classifier
Market Data Providers Yahoo Finance (yfinance), NSE Data Stream Real-time & Intraday
Persistence & Cache SharedPreferences, SQLite Local Storage Instant Load Times

🏛️ System Architecture

flowchart TD
    subgraph Client ["Flutter Cross-Platform Client (Web, Android, Windows)"]
        UI["UI Screens (Home, Market, Stock Detail, Copilot, Portfolio)"]
        Riverpod["Riverpod 2.x State Management"]
        Router["GoRouter Declarative Navigation"]
        Charts["Syncfusion Candlesticks & Brownian Bridge Engine"]
        Storage["StorageService (Encrypted Vault & Paper Ledger)"]
        Shield["SecurityLifecycleWrapper (App Switcher Privacy Shield)"]
    end

    subgraph Backend ["FastAPI High-Performance Engine (Python 3.10+)"]
        API["FastAPI REST & WebSocket Controllers"]
        Orchestrator["AI Orchestrator & Intent Router"]
        TechEngine["Technical Analysis Service (RSI, MACD, BB)"]
        MLService["Universal XGBoost v2 Prediction Service"]
        NewsEngine["Live News & FinBERT Sentiment Engine"]
        ReportService["Grounded Research Report Generator"]
    end

    subgraph Intelligence ["External Intelligence & Data Sources"]
        YF[("Yahoo Finance & NSE Data Streams")]
        Groq[("Groq Cloud (Llama 3 70B @ 300+ T/s)")]
        FinBERT[("FinBERT NLP Financial Sentiment")]
    end

    UI --> Riverpod
    Riverpod --> Charts
    Riverpod --> Storage
    Riverpod --> Router
    Riverpod --> Shield
    Riverpod <-->|REST & WebSocket| API

    API --> Orchestrator
    API --> TechEngine
    API --> MLService
    API --> NewsEngine
    API --> ReportService

    TechEngine --> YF
    MLService --> TechEngine
    NewsEngine --> FinBERT
    Orchestrator --> Groq
    Orchestrator --> MLService
    Orchestrator --> TechEngine
    Orchestrator --> NewsEngine
Loading

🧠 Universal XGBoost v2 Machine Learning Engine

The core machine learning engine in TradeVision is a Universal Gradient Boosted Decision Tree (XGBoost v2) model trained on normalized technical indicators and microstructural price action across the NSE universe.

📊 Feature Engineering Matrix (27 Features)

Category Features Description
Momentum rsi_14, mom_1d, mom_5d, mom_10d, mom_21d Rate of change across multiple time horizons and RSI momentum oscillator
Trend Following ema_9_dist, ema_21_dist, ema_50_dist, ema_200_dist, trend_strength Percentage distance from key exponential moving averages and composite trend direction
MACD Oscillators macd_line, macd_signal, macd_hist, macd_slope Moving average convergence/divergence and momentum acceleration
Volatility bb_pct_b, bb_bandwidth, atr_14, historical_vol_21 Normalized Bollinger Band position, band squeeze metrics, and Average True Range
Volume Dynamics vol_ratio_20, obv_slope_5, vwap_deviation Relative volume compared to 20-day average, On-Balance Volume slope, and VWAP delta
Price Action high_low_ratio, body_wick_ratio, candle_direction Intraday candle geometry, upper/lower wick rejection, and candle polarity
Market Microstructure day_of_week, session_phase Temporal and session-opening/closing microstructural adjustments

🎯 Empirical Model Performance

  ╔══════════════════════════════════════════╦═════════════════════════╗
  ║ Evaluation Metric                        ║ Benchmark Score         ║
  ╠══════════════════════════════════════════╬═════════════════════════╣
  ║ Overall Test Set Accuracy                ║ 84.62%                  ║
  ║ Calibrated High-Confidence Gating        ║ 97.19% Accuracy         ║
  ║ Area Under ROC Curve (ROC-AUC)           ║ 0.9869                  ║
  ║ Precision (Bullish Outperform)           ║ 0.9680                  ║
  ║ Recall (Directional Detection)           ║ 0.9540                  ║
  ║ Mean Inference Latency                   ║ < 12 ms                 ║
  ╚══════════════════════════════════════════╩═════════════════════════╝
📐 View Mathematical Gating Formulation

The confidence gate uses a calibrated sigmoid boundary function:

$$\text{Gate}(P) = \begin{cases} \text{BUY}, & \text{if } P(\text{Bull}) \ge 0.55 + \delta \ \text{SELL}, & \text{if } P(\text{Bull}) \le 0.45 - \delta \ \text{HOLD / CONSOLIDATION}, & \text{otherwise} \end{cases}$$

Where $\delta$ represents the volatility-adjusted buffer. When the probability falls within the uncertainty band $[0.45, 0.55]$, the model refuses to guess, filtering out rangebound whipsaw trades.


🤖 Groq Llama 3 70B Copilot

TradeVision AI integrates Groq Cloud's ultra-low latency LPU inference engine running Meta's llama-3.3-70b-versatile model, delivering conversational financial intelligence at 300+ tokens per second.

                    User Query / Stock Analysis Request
                                    │
                                    ▼
                     [AI Orchestrator Service]
                                    │
           ┌────────────────────────┼────────────────────────┐
           ▼                        ▼                        ▼
    [App Navigation &       [Stock Technical        [Market Macro &
     Personal Greeting]      & ML Signal]            Live News Feed]
           │                        │                        │
           └────────────────────────┬────────────────────────┘
                                    │
                                    ▼
                     [Depth Scaling Controller]
            ┌───────────────────────┼───────────────────────┐
            ▼                       ▼                       ▼
     [Quick Mode]            [Standard Mode]         [Deep Mode]
      250 Tokens              650 Tokens              1500 Tokens
      Quick levels & bias     Full indicators         Institutional target,
                              & rationale             SL, and scenarios
                                    │
                                    ▼
                     [Formatting Post-Processor]
             (Strips raw hash headers & broken table pipes)
                                    │
                                    ▼
                         Rendered in Flutter UI

🎓 Academic Research & Viva Defense

This project has been developed as an engineering capstone and research endeavor at Department of Computer Engineering, Shri Bhagubhai Mafatlal Polytechnic & College of Engineering, Mumbai, under the mentorship of Prof. Prachi Arora (Senior Lecturer).

📑 Key Project Documentation


📂 Repository Structure

TradeVision/
├── docs/                                  # Academic & Defense Documentation
│   ├── VIVA_DEFENSE_GUIDE.md              # 2-Min Pitch, Top 10 Q&A & Examiner Protocol
│   └── RESEARCH_PAPER.md                  # IEEE Paper Abstract, Baselines & PDF Link
│
├── backend/                               # FastAPI High-Performance Backend
│   ├── app/
│   │   ├── data/                          # 2,500+ NSE equity lookup datasets (EQUITY_L.csv)
│   │   ├── models/                        # Serialized ML weights & metadata
│   │   ├── routers/                       # REST endpoint controllers (Auth, Stocks, ML, AI)
│   │   ├── schemas/                       # Pydantic v2 validation models
│   │   ├── services/                      # AI orchestrator, ML predictor, Groq & News
│   │   └── main.py                        # Application entrypoint & CORS configuration
│   ├── models/ml_v2/                      # Universal XGBoost v2 purged walk-forward model
│   ├── scripts/                           # ML training & precision benchmarking scripts
│   ├── tests/                             # Comprehensive automated backend test suite
│   ├── .env.example                       # Clean backend environment variable template
│   └── requirements.txt                   # Pinned Python package dependencies
│
├── frontend/                              # Flutter Cross-Platform Client
│   ├── android/                           # Android native configuration (Java 17, Desugaring)
│   │   └── app/src/main/res/drawable/     # Native ic_notification.png status bar icon
│   ├── assets/                            # Bundled stock universe JSON & vector brandmarks
│   ├── lib/
│   │   ├── core/                          # Design tokens, security lifecycle & error handlers
│   │   ├── models/                        # Client-side Dart data models
│   │   ├── providers/                     # Riverpod state providers
│   │   ├── router/                        # GoRouter navigation configuration
│   │   ├── screens/                       # Presentation screens (Home, Market, Charts, Copilot)
│   │   ├── services/                      # API client, WebSockets & StorageService
│   │   ├── widgets/                       # Reusable UI components & 3D flip card engine
│   │   └── main.dart                      # Flutter client entrypoint
│   └── pubspec.yaml                       # Dart dependencies & asset declarations
│
├── start_presentation.bat                 # 1-Click Windows Batch Demo Launcher
├── start_presentation.ps1                 # 1-Click PowerShell Demo Launcher
├── sync_repos.ps1                         # Dual-repo synchronization script
├── TradeVision_AI_Final.apk               # Final Production Android APK (v1.0.0+2)
├── .gitignore                             # Strict security & document exclusion rules
└── README.md                              # Master repository documentation

🚀 Quick Start

⚡ Option A: 1-Click Instant Demo Launcher (Fastest for Viva & Review)

If you are on Windows, simply run the bundled launcher to start both Backend and Frontend in seconds:

# In PowerShell:
.\start_presentation.ps1

# OR simply double-click:
start_presentation.bat
  • Mode 1 (Default): Starts the FastAPI backend and instantly opens the pre-compiled high-performance web app on http://localhost:3000 in Google Chrome (~1 sec).
  • Mode 2: Launches Flutter Dev Server with hot-reload.
  • Mode 3: Opens interactive Swagger API documentation on http://127.0.0.1:8000/docs.

🛠️ Option B: Manual Step-by-Step Setup

Prerequisites

1️⃣ Backend Setup

cd backend
python -m venv venv

# Activate Virtual Environment (Windows PowerShell):
.\venv\Scripts\Activate.ps1
# macOS / Linux: source venv/bin/activate

pip install -r requirements.txt

# Copy environment template and add your Groq key
copy .env.example .env

# Launch FastAPI server
uvicorn app.main:app --host 127.0.0.1 --port 8000 --reload

Swagger UI is immediately available at http://127.0.0.1:8000/docs.

2️⃣ Frontend Setup

cd ../frontend
flutter pub get

# Run on Chrome
flutter run -d chrome --web-port 3000

# Run on Android Device / Emulator
flutter run -d android

📱 3️⃣ Build & Install Android Final Release APK

The project includes precompiled universal binaries:

  • Direct Path: TradeVision_AI_Final.apk (Version 1.0.0+2, 74.78 MB)

To compile a fresh release from source:

cd frontend
flutter build apk --release

The newly compiled APK is output to: frontend/build/app/outputs/flutter-apk/app-release.apk


📡 REST & WebSocket API Reference

Method Endpoint Description
GET /api/v1/stocks Real-time quotes for benchmark Indian equities
GET /api/v1/stocks/{ticker} Fundamental ratios, 52-week high/low, P/E, market cap
GET /api/v1/chart/{ticker}?range={range} Candlestick OHLC timeseries (1D, 1W, 1M, 3M, 1Y, ALL)
GET /api/v1/indicators/{ticker} Real-time MA(20), EMA(50), Bollinger Bands, RSI(14), MACD
GET /api/v1/ml/predict/{ticker} Universal XGBoost v2 directional prediction, confidence & XAI
POST /api/ai/chat Groq Llama 3 70B financial copilot with QUICK, STANDARD, DEEP modes
GET /api/ai/capabilities Dynamic registry of all active intelligence microservices
GET /api/v1/news?ticker={ticker} Real-time financial headlines with FinBERT sentiment scores
WS /ws/market-stream High-frequency live tick stream for active watchlists

🗺️ Product Roadmap

  • Universal XGBoost v2 ML Engine: 27 engineered technical & microstructural features with calibrated gating.
  • Groq Llama 3 70B Copilot: 3 depth modes (Quick, Standard, Deep) with conversational intent routing.
  • Interactive Virtual Paper Trading: ₹10,00,000 capital account, instant order execution, and live P&L.
  • Syncfusion Candlestick Charts & Brownian Bridge: Pinned stochastic interpolation with RSI, MACD, and BB.
  • Institutional Security Layer: App switcher blur shield, encrypted vault storage, and native notification icons.
  • 1-Click Viva & Demo Launcher: Instant cross-platform demonstration utility.
  • FinBERT Live Sentiment Engine: Real-time financial headline scraping and polarity scoring.
  • Live Indian Standard Time (IST) Engine: Automatic trading session phase detection.
  • Level 2 Real-Time Order Book: Visual bid/ask market depth ladder with live participant weighting.
  • Algorithmic Strategy Backtesting: Custom strategy builder with historical equity replay.
  • Voice-Enabled Market Copilot: Natural language voice queries for instant stock screening.

⚠️ Regulatory & Educational Disclaimer

STATUTORY DISCLAIMER (SEBI Compliance):
TradeVision AI is an educational and academic research platform developed for demonstration purposes. The analytical outputs, machine learning predictions, and AI-generated signals provided by this software do NOT constitute financial, investment, legal, or trading advice under SEBI (Securities and Exchange Board of India) regulations or any other jurisdictional framework. Stock markets involve substantial risk of loss. Always consult a certified financial advisor before committing real capital.


👥 Contributors


Footer

⭐ Star us on GitHub if you find TradeVision AI inspiring!

Built with ❤️, Flutter, FastAPI, and Advanced Artificial Intelligence.

About

AI-driven equity intelligence platform featuring Brownian Bridge chart dynamics, FinBERT sentiment analysis, ML-powered BUY/SELL signals, and real-time market insights — built with Flutter & FastAPI.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages