CoinCard

CoinCard transforms crypto investing into an autonomous, insight-driven experience. By combining live market data, X trends, and AI reasoning, it guides users to buy, sell, or hold confidently.

Comment

CoinCard - Coinbase Replica

Members

Inspiration

Traditional trading dashboards dump charts, news, and tickers onto one page and leave the user to stitch the story together. We wanted CoinCard to feel like an AI-native version of Coinbase: the app should understand the news it pulls, reason about sentiment, know how a specific user trades, and proactively draft trades the user can accept or ignore.

What it does

CoinCard transforms crypto investing into an autonomous, insight-driven experience. By combining live market data, X (Twitter) trends, and AI reasoning, it guides users to buy, sell, or hold confidently.

Backend Features

  • Fetches real-time crypto data (auto-refreshes every 10 minutes)
  • Provides visual analytics for each coin
  • Tracks live crypto trends and community sentiment from X (Twitter)
    • Use X API to fetch all tweets related to a coin
    • Integrate xAI’s Sentiment Model to classify emotions: BULLISH, NEUTRAL, or BEARISH
    • ⇒ Help users spot emerging trends before price movements
  • Enable users to BUY, SELL, and CONVERT coins seamlessly through a unified panel

AI System

  • AI Chat Assistant (“Ask Grok”): Embedded everywhere in CoinCard from Home to Trade tabs. Use xAI API (Grok) with custom Prompt Engineering to give users more understanding of a specific coin.
  • Strategic Lab: Orchestrate multiple agents (News, Sentiment, Behavioral, Analysis, VibeTrade) using LangGraph
  • Style Match: Use RAG to learn user’s investing style from coins they’ve bought, coins they’ve asked the chatbot about, and past portfolio performance and preferences
  • Anomaly Detection (Machine Learning): Powered by Snowflake Cortex Anomaly Detection, trained directly on live crypto price data

How we built it

  • Backend: FastAPI, Snowflake, Grok (xAI). Each agent exposes its own router (news, behavioral, live_trade, orchestrator) so the frontend can compose flows freely.
  • Frontend: React + Vite + TypeScript, split between a protected dashboard shell and multiple views (Home, Live Trading, Strategic Lab). We stream orchestration steps via SSE so users see Grok’s thought process as it happens.
  • Data/AI: Grok handles summarization, reasoning, orchestration, and the live trade agent. CoinGecko supplies historical prices. Sentiment scores come from the news agent’s per-doc inference.

Challenges

  • A member had no idea about trading or crypto before so we struggled teaching him
  • Only 1 member knew how to use Snowflake before so it was difficult for the others to catch up with a large-scale database (of course, we're working on a finance project)
  • Designing prompts so Grok emits strict JSON for trade plans without hallucinating missing fields
  • Keeping streaming UI responsive while FastAPI streamed SSE; we had to build JSONL and SSE branches to satisfy both Strategic Lab and Live Desk
  • Aligning behavioral analysis (Snowflake transactions) with recommended coins, then merging those results into the Strategic Lab orchestration without coupling it to the live trade agent

Accomplishments

  • Use Snowflake API thoroughly from database, Streams & Tasks, snowpark, to live SQL queries
  • Built a transparent orchestration log that shows every tool call, candidate tool scores, and final trade plan
  • Created a live trading flow where sentiment, price action, and structured deals update automatically on token change
  • Delivered consistent UX patterns (sentiment tags, deal cards, action buttons) across pages so AI suggestions feel native to the app

What we learned

  • LangGraph
  • Snowflake SQL connection between backend (Python) and database
  • Trading! It's so fun and risky!

What’s next

  • Connect /orchestrate/execute to real brokerage/order rails with safety checks and paper-trading first.
  • Extend the behavioral agent to learn user preferences over time and auto-adjust the live trade agent’s risk levels.
  • Add portfolio hedging simulations and “what-if” toggles inside Strategic Lab.
  • Polish authentication and multi-user tenancy so each trader can save plans, live decisions, and execution logs.

Running locally

  1. Backend

    cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
uvicorn src.main:app --reload

    Configure .env with XAI_API_KEY, Snowflake credentials, CoinGecko (optional), etc.

  2. Frontend

    cd frontend
npm install
npm run dev

    Visit http://localhost:3000, log in with a test user, and explore Home, Live Trading, and Strategic Lab.

Built with

  • FastAPI, Python
  • React, TypeScript
  • X API, CoinGecko API
  • xAI API (Grok)
  • Snowflake API
    • Snowflake Database
    • snowflake.connector
    • snowpark
    • Snowflake Cortex Anomaly Detection
    • Snowflake Streams & Tasks

Built With

Share this project:

Updates