NeuroTrade AI: Semantic Trader Insights with MongoDB

We turn trading data into real-time trader intelligence using AI embeddings and MongoDB Vector Search, enabling smarter strategies and behavioral feedback from your own trading patterns.

Comment

Inspiration

As a trader and software engineer, I’ve always been frustrated by how trading platforms only show performance metrics profits, losses, and charts but never the why. I wanted to build a system that goes deeper. One that understands trading behavior using AI, and helps people make smarter, more reflective decisions. MongoDB’s flexible schema and Vector Search capabilities made this possible.

What it does

NeuroTrade connects to a trader’s OANDA account and pulls live data about trades, account metrics, and candle charts. It embeds that data using either OpenAI or Vertex AI, then stores the vectors and metadata in MongoDB Atlas.

Traders can then ask natural-language questions like:

  • “Which trades had similar behavior?”
  • “When did I repeat this mistake?”
  • “What trade setups worked best last week?”

The system finds similar embeddings using MongoDB’s vector search, and returns AI-generated insights in real-time.

How we built it

I broke the application into five GitLab repositories:

  • neurotrade_frontend (React)
  • neurotrade_backend_service (routing and auth)
  • neurotrade_oanda_service (trade data ingestion)
  • neurotrade_oanda_sync (historical data ingestion since 2005)
  • neurotrade_openai_service and neurotrade_vertexai_service (embedding / query processors)

Data is sent to embedding services via queues (RabbitMQ), then stored in MongoDB using collections like trades, accounts, candles, and embeddings. Queries go through semantic search using MongoDB Vector Search, returning contextual insights.

Challenges we ran into

  • Normalizing inconsistent trade formats from OANDA
  • Structuring metadata for accurate vector similarity
  • Managing embedding costs and token limits efficiently
  • Ensuring query relevance without losing performance
  • Maintaining service coordination across 5 microservices

Accomplishments that we're proud of

  • Built a fully functional behavioral intelligence engine for traders in just a few weeks
  • Implemented live semantic search on embedded trade data using MongoDB Atlas
  • Successfully embedded over 1,000 trade events with accurate similarity results
  • Deployed a modular microservice architecture with full GitLab CI/CD automation
  • Enabled real users to create accounts and run trade behavior queries through a live web app

What we learned

  • MongoDB Atlas Vector Search is ideal for real-time AI + data search use cases
  • Embeddings are only as valuable as the metadata you structure around them
  • Working with queues and multiple AI pipelines requires solid service orchestration
  • Traders are eager for insights but those insights must be fast, clear, and personal

What's next for NeuroTrade AI: Semantic Trader Insights with MongoDB

  • Add more broker integrations beyond OANDA (e.g., MT4, Deriv, TradingView)
  • Expand behavioral scoring to include risk exposure, timing quality, and emotion detection
  • Build a mobile companion app for traders to get feedback while on the go
  • Create a “replay mode” where traders can rewalk through past trading sessions with AI
  • Offer a visual explorer powered by MongoDB aggregation pipelines and heatmaps

Built With

Share this project:

Updates