NanoMech

NanoMech overlay in standby mode — sitting transparently over Binance chart ready to scan
NanoMech scanning in Auto Mode — sending screenshot to Gemini 2.5 Flash on Google Cloud
NanoMech showing full market analysis — Trend, Liquidity, Momentum with Trade Setup and Risk Calculator
NanoMech system architecture — Screen capture to Gemini 2.5 Flash on Google Cloud

Inspiration

Traders waste precious seconds switching between their chart, analysis tools, and risk calculators. By the time the setup is ready, the candle has moved. I wanted to build an AI that sits on top of your screen, sees exactly what you see, and gives you the full trade plan in seconds — without ever leaving your chart.

What it does

NanoMech is a real-time AI trading assistant that runs as two transparent overlays on top of any trading platform. It uses Google Gemini 2.5 Flash multimodal vision to read your chart visually and deliver a complete trade plan instantly.

Overlay 1 — Market Analysis:

Trend — bullish/bearish structure, moving average crossovers
Liquidity — order book depth, bid/ask walls, support/resistance zones
Momentum — candlestick patterns, volume behavior, price velocity

Overlay 2 — Trade Setup + Risk Calculator:

Entry Price, Target Price, Stop Loss price
CALC section: Risk Amount ($), Position Size (Units), R:R Ratio — all update live as you type your risk %

How we built it

Captured the screen in real-time using mss library
Sent the screenshot to Google Gemini 2.5 Flash via Google GenAI SDK on Google Cloud
Engineered structured prompts using [ANALYSIS] and [TRADE] section tags for reliable parsing
Extracted Entry, Target, Stop prices using regex and wired them into the risk calculator
Built two frameless always-on-top transparent overlays using Tkinter
Used threading to keep UI responsive during API calls with a spinning animation
Added global hotkey Ctrl+A+I and Auto Mode for hands-free scanning every 20 seconds

Challenges we ran into

Getting Gemini to consistently return prices in a parseable numeric format required careful prompt engineering and fallback handling
Tkinter is not thread-safe so all UI updates had to be routed through root.after() callbacks from the analysis thread
Tuning overlay transparency and colors to stay readable across both dark and light chart themes
Global hotkeys conflicting with trading platform shortcuts

Accomplishments that we're proud of

Built a fully working multimodal AI agent that reads ANY chart on ANY platform with no exchange API
Seamless overlay UX that never interrupts the trader's workflow
Complete risk management calculator driven entirely by AI-extracted price levels
Clean structured outputs from Gemini through prompt engineering alone

What we learned

How capable Gemini 2.5 Flash is at reading financial charts visually — it correctly identifies candlestick patterns, moving averages, volume spikes, and price levels from a raw screenshot
Prompt engineering for structured and reliably parseable AI output
Building production-grade always-on-top desktop overlays in Python
How to handle threading safely in Tkinter applications

What's next for NanoMech

Voice output — NanoMech speaks the trade setup aloud for fully hands-free trading
Multi-monitor support — choose which screen to analyze
Trade history tracker — log and measure how AI setups performed over time
Cloud hosted version — run NanoMech as a web service on Google Cloud Run