Built at Hacklytics 2026: Golden Byte by Risheet Jajoo, Vikram Ren, Shashank Shaga, and Sai R.

Over the past year, escalating tariff announcements have triggered immediate and often severe volatility across global equity markets. Traditional financial models assume smooth price evolution, while geopolitical policy moves create sudden structural shocks. Prediction markets often reflect sentiment shifts before institutions fully price them in — yet no existing tool quantifies how tariff data affects market trends across countries and sectors.

Our core question: What if we could quantify tariff uncertainty before it hits and simulate its ripple effects across global markets in real time?

Quantara was built to transform policy uncertainty into measurable, forward-looking financial insight.

Quantara is a full-stack predictive analytics platform that stress-tests global markets against shifting tariff landscapes. By bridging prediction market sentiment with stochastic financial modeling, it provides a real-time macroeconomic impact simulator.

Key capabilities:

• Calculates a "Surprise Gap" between market expectations and tariff probability
• Visualizes geopolitical risk via an interactive global choropleth map
• Simulates 90-day price trajectories for major indices and sectors
• Models discontinuous policy shocks using jump-diffusion mathematics
• Translates complex quantitative outputs into plain-English via an AI analyst

Instead of reacting after the market moves, Quantara enables proactive macro risk assessment.

Quantara operates through a dual-engine architecture supported by an intelligence layer.

The core of Quantara's probability estimation is a stacked ensemble model that combines CatBoost and Logistic Regression to predict the likelihood of a tariff event occurring for a given country or sector.

Model Architecture:

• CatBoost serves as the primary learner, handling nonlinear feature interactions and categorical geopolitical variables without requiring extensive preprocessing.
• Logistic Regression acts as a calibrated second-stage model, taking CatBoost's output probabilities alongside raw features to produce a well-calibrated final tariff probability.
• The stacked output is then combined with Kalshi prediction market data to compute the Surprise Gap — the delta between what our model predicts and what the market has already priced in.

Training Datasets:

Feature Engineering highlights:

• Country-level trade deficit and surplus ratios
• Rolling news sentiment scores leading up to policy events
• Historical tariff frequency and escalation patterns
• Macroeconomic stress indicators (e.g. trade-to-GDP ratio)

The ensemble probability output is combined with Kalshi market-implied probabilities to produce the Surprise Gap — a forward-looking policy stress indicator surfaced through an interactive global choropleth map that lets users instantly identify geopolitical hotspots.

We convert the Surprise Gap into projected financial impact using a Merton Jump-Diffusion model. Unlike traditional geometric Brownian motion models, this framework captures:

• Jump Component: The instantaneous market shock following a tariff announcement
• Diffusion Component: The longer-term cost absorption and repricing process

We run Monte Carlo simulations to generate 90-day probabilistic trajectories for:

• Major indices (S&P 500, Nasdaq, Dow)
• Sector-specific exposure
• Country-sensitive industries (e.g., automotive, semiconductors)

A Retrieval-Augmented Generation (RAG) chatbot powered by Snowflake Cortex interprets complex outputs in plain English. The AI analyst:

• Interprets jagged simulation graphs
• Correlates projections with policy documents
• Connects outcomes to supply chain metadata
• Explains sector-specific valuation cliffs clearly

Hacklytics-2026/
├── Triumph/            # Frontend application
├── backend/            # API and data pipeline
├── chatbot/            # RAG chatbot integration (Snowflake Cortex)
├── ml/                 # Stochastic modeling and Monte Carlo engine
├── Trained_models/     # Saved model outputs
└── train_1.py          # Model training entrypoint

• Node.js 18+
• Python 3.9+
• Snowflake account (for RAG chatbot)
• Kalshi API key

# Clone the repository
git clone https://github.com/rjajoo18/Hacklytics-2026.git
cd Hacklytics-2026

# Install frontend/backend dependencies
npm install

# Install Python dependencies
pip install -r ml/requirements.txt   # (or install yfinance, numpy, scipy, etc.)

# Set up environment variables
cp .env.local.example .env.local
# Fill in your API keys (Kalshi, Snowflake, etc.)

# Start the backend
cd backend && npm run dev

# Start the frontend (in a separate terminal)
cd Triumph && npm run dev

# Run the ML simulation engine
python ml/simulate.py

• Translating qualitative geopolitical uncertainty into a measurable numerical shock variable
• Calibrating jump intensity and diffusion parameters without overfitting
• Aligning prediction market probabilities with financial volatility data
• Maintaining interpretability while preserving quantitative rigor
• Integrating real-time data pipelines within a hackathon timeline

• Built and trained a stacked ensemble (CatBoost + Logistic Regression) to predict tariff probabilities from real-world macroeconomic, sentiment, and policy data
• Successfully fused ML model outputs with Kalshi prediction market data to compute a novel Surprise Gap metric
• Implemented a working Merton Jump-Diffusion simulation engine fed directly by the Surprise Gap
• Built an interactive geopolitical risk visualization interface
• Deployed a RAG-powered AI analyst for contextual interpretation
• Delivered a cohesive full-stack macro stress-testing platform within a hackathon timeframe

• Expand beyond tariffs into sanctions and monetary policy shocks
• Integrate additional macroeconomic and prediction data sources
• Add portfolio-level stress testing tools
• Deploy live APIs for institutional use
• Enable customizable scenario simulations for advanced users

Long-term goal: Make macroeconomic shock modeling proactive, interpretable, and accessible — not reactive.

This project was built for Hacklytics 2026. Feel free to explore and build on it!