PulseTrade AI: The Live Crypto Commentator

PulseTrade AI turns crypto market chaos into live sports commentary.Powered by Gemini and ElevenLabs, it detects realtime anomalies and narrates the action so you can listen to the market's heartbeat.

Comment

PulseTrade AI -- Live Crypto Commentator

Inspiration

The idea for PulseTrade AI was born out of frustration with traditional stock market alerts. Initially, we built a standard stock tracker using Yahoo Finance, but we quickly hit a wall:

Markets don’t wait for API polls.

Polling data every 5 seconds meant we were always looking at the past. By the time an alert triggered, the move was already over.

To truly capture the pulse of the market, we realized we needed:

  • Real-time precision -- millisecond-level data, not 5-second snapshots
  • Immersive experience -- instead of staring at charts, what if the market could speak to you like a sports commentator?

This insight led to a strategic pivot from traditional stocks to Crypto (Binance WebSockets) and the birth of the Live AI Commentator concept.

What It Does

PulseTrade AI turns crypto market chaos into a live sports broadcast.

It connects directly to the Binance exchange, listening to every trade for major assets like Bitcoin and Solana. Instead of showing a green candle on a chart, it detects anomalies:

  • Sudden volume spikes
  • Whale movements
  • Psychological level breaks

When something meaningful happens, it instantly generates voice-narrated commentary.

It’s not just a dashboard - it’s a 24/7 financial radio station generated entirely by AI in real time.

How We Built It

We engineered a pipeline optimized for speed, splitting the system into distinct “organs” aligned with the hackathon’s core technologies.

The Ears (Ingestion)

  • A Python service connects continuously to the Binance WebSocket
  • Ingests raw trade data for BTC, ETH, and SOL

The Nervous System (Confluent)

  • Confluent Cloud (Apache Kafka) buffers the high-velocity trade stream
  • Ensures zero data loss, even if downstream AI processing lags
  • Acts as a shock absorber for financial-grade reliability

The Brain (Analysis & AI)

Technical Engine

  • Real-time indicators like RSI and Rolling Volume
  • RSI is computed using the standard formula:

[ RSI = 100 - \frac{100}{1 + RS} ]

Narrative AI

  • When a signal triggers (e.g., Volume > 5 × μ_vol)
  • Context is sent to Gemini 2.5 Flash
  • Generates a one-sentence high-energy market commentary

The Voice (ElevenLabs)

  • Commentary is streamed to ElevenLabs Turbo v2.5
  • Chosen for ultra-low latency
  • Ensures audio lands while the candle is still green

The Guardian (Datadog)

  • Entire backend wrapped with Datadog APM
  • Full visibility into “Time to Voice” latency
  • Enabled shaving milliseconds off the critical path

Challenges We Ran Into

The “Chatty AI” Problem

  • Early versions reacted to every micro-movement
  • Solution: a Breakout Router to filter noise
  • AI only speaks on significant events
    • Example: Price move > 1% in 60s

Audio Latency

  • Audio generation naturally introduces delay
  • Optimized by:
    • Using streaming responses from ElevenLabs
    • Playing audio chunks immediately on the frontend

Browser Autoplay Policies

  • Browsers block autoplay audio
  • Solved with a Click-to-Enable UI pattern
  • Introduced the “Audio Pulse” button to safely initialize audio context

Accomplishments We’re Proud Of

  • Sub-2 Second Latency
    AI commentary arrives as the pump happens, not after

  • Full Observability
    Distributed traces across containers via Datadog gave deep confidence in system stability

  • Premium Experience

A polished dark-mode Bento Grid UI, TradingView charts, and reactive visualizers
Feels like professional trading software, not a hackathon demo

What We Learned

  • Event-Driven > Polling
    Stop asking “What is the price?”
    Start listening for “The price just changed.”

  • The Power of Sound
    Audio creates urgency that visuals simply can’t match

  • Microservices at Scale
    Kafka allowed us to decouple:

    • High-velocity ingestion (100+ trades/sec)
    • Heavy AI processing
      Without dropping data

What’s Next: The Live Crypto Commentator

Personalities

  • Bullish Degen
  • Bearish Skeptic
  • Professional Analyst
    Users choose who narrates their market

Live Trading

  • Wallet integrations
  • Execute trades with a single click (or voice command) after an alert

Mobile App

  • iOS & Android support
  • Listen to the market while driving, working out, or multitasking

PulseTrade AI

Hear the market. Trade the moment.

Built With

  • binance-websocket-api-observability:-datadog-apm
  • datadog-tracing-tools:-shadcn-ui
  • docker-ai-&-apis:-gemini-1.5-flash-(google-vertex-ai)
  • elevenlabs-turbo-v2.5
  • fastapi
  • languages:-python
  • sql-frameworks:-next.js-14
  • tailwind-css-cloud-&-infrastructure:-confluent-cloud-(apache-kafka)
  • tradingview-lightweight-charts
  • typescript
  • websockets
Share this project:

Updates

Submission history