Quantwater Cognitive Terminal

Quantwater Cognitive Terminal is an AI-native macro workstation that fuses market data with Gemini 3 Flash to convert raw time-series into institutional-grade alpha intelligence.

Comment

Inspiration

The gap between raw data and alpha is widening. While most retail platforms provide modest access, they lack the reasoning capabilities of high-end institutional systems like Bloomberg or Reuters. Inspired by recent breakthroughs in Agentic Vision, we engineered Quantwater to act as a synthetic analyst. Our vision was to combine the excellent accessibility of platforms like MT5 with the cognitive power of Gemini 3, creating a terminal that doesn't just show the chart, but reads it.

What it does

Quantwater Cognitive Terminal is a professional multi-asset workstation providing:

  • F1: Equity Analysis: Real-time deep dives into individual stocks with AI-augmented technical indicators.
  • F2: Macro Dashboard: A "Ground Truth" view of global liquidity, tracking 10Y-2Y yield spreads and synthetic Dollar Index ($DXY) moves.
  • F3: Global Monitor: A synchronized 2x3 grid for identifying capital rotation across Equities, FX, Rates, and Commodities.
  • Agentic Vision: A "Hero Feature" where the AI visually scans 15-minute charts to identify structural support, resistance, and complex patterns like "rounding bottoms" using Python code execution for pixel-perfect accuracy.

How we built it

We engineered a "Causal Data Lake" pipeline:

  1. Ingestion: Proprietary MetaTrader 5 scripts feed raw M15 bars into the lake.
  2. Silver Layer: A high-performance DuckDB database that calculates institutional yields and cross-asset correlations in sub-10ms.
  3. The Bridge: A FastAPI backend that acts as the "Nervous System," providing cached AI research and time-series data.
  4. Frontend: A high-density React 19 UI built for low-glare trading environments.
  5. Intelligence: Gemini 3 Flash serves as the Lead Strategist, utilizing Thinking Level: High for macro notes and Agentic Vision for visual technical analysis.

Challenges we ran into

  • Memory Constraints: Optimizing a high-density 2x3 grid of live charts on an 8GB RAM workstation required refactoring React state logic and disabling expensive animations.
  • Bond Proxy Accuracy: Converting ETF prices into professional yields required a calibrated duration-based mathematical model.
  • API Quota Management: We solved 429 errors by implementing a server-side caching layer and a manual "Strategic Refresh" trigger.

Accomplishments that we're proud of

We are most proud of the Agentic Vision implementation. Watching the AI identify a "liquidity flush" on a Gold chart by executing Python code on the visual pixels—just like a human trader—was our "Aha!" moment. We also achieved 99% accuracy in our bond yield estimates compared to real-world treasury rates.

What we learned

Building Quantwater taught us that Gemini 3 is no longer just a chatbot; it is a reasoning engine. We learned how to "ground" an LLM in a private data lake to prevent hallucinations, turning it into a reliable institutional-grade strategist.

What's next for Quantwater Cognitive Terminal

  • Predictive ML Layer: Integrating XGBoost models to predict "Liquidity Gaps" before they happen.
  • Trade Execution: Connecting the "Vision Eye" directly to MT5 order routing for one-click agentic trading.
  • Expanded Universe: Adding emerging market data and alternative data sets like sentiment from social feeds.
  1. Video Link: https://youtu.be/St3TQP5hrhk
  2. Repo Link: https://github.com/pawan-pro/CLI-Finance-Terminal/tree/submission-final
  3. Built With: Gemini API, React, Python, FastAPI, DuckDB.

Built With

Share this project:

Updates