Live Neural Volatility Surface

Your AI options
trading co-pilot

Talk to your volatility surface. Draw custom price paths. Get instant breakeven analysis.
Powered by a physics-informed neural network that beats the 30-year industry standard.

0%
Better than SABR 30-year industry gold standard
Scroll
0
Neural Parameters
0
Training Time
0
Parametric Assumptions
<5ms
Per-point Latency
0
Better than SABR

Meet your AI
trading assistant

Powered by ElevenLabs voice + Gemini intelligence. Ask for option prices, run breakeven analysis, and command the AI to draw custom market scenarios — all by voice.

🎤

Voice-driven option pricing

Ask "What is the SPY 580 call expiring Friday worth?" and get instant live pricing from the neural surface.

📈

Draw scenarios by voice

Say "Draw a V-recovery to $700 then crash" and watch the canvas populate your scenario in real time.

⚖️

Instant breakeven analysis

Ask what price SPY needs to be at for your option to break even, accounting for time and vol crush.

Query live training status

Ask if the neural network is trained, check MSE against SABR, or trigger a fresh training run.

Σ
Sigma
Live — Neural surface loaded
Σ
Neural surface trained. SPY at $679.46 — 843 contracts loaded across 13 expirations. What do you want to know?
U
What's the SPY 680 call expiring this Friday worth right now?
Σ
SPY 680C (3 DTE) from the neural surface:
Price $4.82
IV 17.3%
Delta 0.48
Theta -$0.31/day
Slightly OTM. If SPY stays flat you lose ~$0.93 by Friday.
U
Draw a V-recovery scenario — crash to 650 then bounce to 700 by Thursday.
Σ
Drawing V-recovery path on canvas now…
Path 679 → 650700
P&L at 700 +$218 per contract
Vol crush -$142 on earnings day
Net P&L +$76 if path holds
The vol crush at the earnings line cuts your gain in half — check the red cliff.
Ask about options, IV, P&L, or say "draw..."
The Problem

Right about the stock.
Lost money anyway.

The market is full of traders who got the direction right and still lost. Nobody explained why — until now.

Stock move
↑ 10%
Earnings beat — stock jumps
Your call option
↓ $4.75
You lost money

Volatility Crush

Before earnings, uncertainty is sky-high. Implied volatility spikes to 60%. Every option buyer is paying a fear premium.

The moment earnings are announced — good or bad — the uncertainty disappears. IV crashes from 60% to 20% instantly. Your option loses more from the vol collapse than it gained from the stock move.

No tool showed you this before you traded. Until now.

"I bought calls on a company that beat earnings by 30%. The stock ripped. My options still lost money. I had no idea this was possible." — The reason Volana exists
How It Works

Train. Draw. See the truth.

Three steps from live market data to a complete picture of your trade — including every scenario your broker doesn't show you.

01

Train the Neural Network

Hit train. Watch a physics-informed neural network learn the entire SPY implied volatility surface from live Yahoo Finance data. ~45 seconds. Zero parametric assumptions.

02
✏️

Draw Your Scenario

Click and drag on the TradingView-style canvas — or just tell the AI to draw it by voice. Set your strike, expiry, and earnings date. The path is your prediction.

03
💡

See the Truth

Your P&L at every point along the path. Green glow = profit. Red glow = loss. The vol crush cliff visible at the earnings line. Ask the AI to explain anything you see.

Features

Everything you need
to see what's really happening

Built for traders who want to understand their options — not just guess.

🌎

3D Neural Vol Surface

Interactive Three.js heightfield showing implied volatility across all strikes and expirations. Rotate, zoom, hover for exact values.

Three.js
✏️

Draw Your Scenario

TradingView-style canvas where you draw price paths by hand or voice. See P&L update in real time as you draw.

Canvas 2D
🎤

AI Voice Agent

Natural conversation about your trades. Ask for prices, breakeven, IV, or command the AI to draw custom scenarios.

ElevenLabs + Gemini
📈

Real-time P&L

Profit and loss calculated at every point along your drawn path using the neural surface — not Black-Scholes approximations.

Live pricing

Vol Crush Simulator

Set an earnings date. See the exact dollar amount you lose when IV collapses — even if your stock prediction is correct.

Earnings edge
🏅

Beats SABR by 43.3%

Zero parametric assumptions. Physics-informed arbitrage constraints. Joint surface fitting across all expirations simultaneously.

PINN model
Head to Head

Neural Net vs SABR

SABR has been the gold standard for vol surface modeling for 30 years. Four parameters, calibrated per expiry. We beat it with zero assumptions.

Feature SABR (30yr standard) Volana Neural Net
Fit method Parametric, per expiry slice Joint surface, 50K params
Assumptions 4 parameters, fixed functional form None
Cross-expiry information Independent slices, no sharing Full surface, information flows
Arbitrage-free By construction (within slice) Physics-informed autograd
Extreme strikes Breaks down — known weakness Smooth extrapolation
Held-out accuracy (MSE) Baseline (0.000091) +43.3% improvement (0.000052)
Vol crush simulation Not supported Built-in simulation
AI voice interface None ElevenLabs + Gemini
Training time Instant (analytic) ~45 seconds (GPU/CPU)
Under the Hood

Neural implicit fields meet
quantitative finance

Same family as NeRF and Gaussian Splatting — learned continuous fields from sparse observations. PINN constraints make no-arbitrage a hard law, not a hope.

Architecture
Input: (K, T) → sin/cos encoding → 32d
Hidden: 4 × 128, SiLU activation
Output: σ (implied vol), Softplus
Params: 50,434 trainable
Loss Function (PINN)
L = L_data // ATM-weighted MSE
  + λL_arb // d²C/dK² ≥ 0, d(σ²T)/dT ≥ 0
  + λL_smooth // d²σ/dK² + d²σ/dT²
// via torch.autograd, create_graph=True
Data Pipeline
Source: Yahoo Finance live option chain
Surface: OTM puts (K<S) + OTM calls (K≥S)
Price: Brent root-finding on bid-ask mid
Clean: Vol>5%, OI>20, IQR per slice
Stack
ML: PyTorch + autograd PINN
Voice: ElevenLabs TTS + Gemini 1.5
API: FastAPI + SSE streaming
3D: Three.js heightfield + raycaster
UI: Vanilla JS + Canvas 2D
Ready?

Train. Draw. Talk.
See the truth.

45 seconds to train. One cursor to draw your scenario.
One voice command to get the answer your broker won't give you.

Launch Volana →
Built in 24 hours at WildHacks 2026 · Northwestern University
Physics-informed neural implicit field · Arbitrage-free by construction
PyTorch FastAPI Three.js ElevenLabs Gemini PINN