Inspiration

Forex trading produces massive amounts of data: prices, trades, account shifts. Most traders only focus on outcomes not behavior. NeuroTrade AI was born from the need to analyze personal trading patterns semantically using AI.

We wanted to build a platform that helps traders reflect, improve, and act using data they already own. GitLab gave us the control, automation, and visibility needed to move fast and stay organized.

What it does

NeuroTrade connects to OANDA, retrieves real-time and historical trade, candle, and account data, and processes it through multiple microservices powered by OpenAI and Vertex AI.

Users can:

  • Ask natural language questions about their trading behavior
  • Search for past trades similar to new scenarios
  • Get AI-generated behavioral insights (e.g., risk exposure, emotional decisions)
  • View semantic trends over time

How we built it

We used a microservice architecture, hosted on an Ubuntu VPS. All services are containerized and deployed through GitLab CI pipelines.

  • GitLab CI/CD: Each microservice repo has its own .gitlab-ci.yml file with test, build, and deploy stages.
  • GitLab Runner: Self-hosted on Contabo, executes Docker jobs per commit.
  • OANDA Service: Fetches raw trade data and sends it to RabbitMQ.
  • OpenAI & Vertex Services: Embed and analyze data using vector models.
  • MongoDB Atlas: Stores raw and embedded data, supports vector search.
  • RabbitMQ: Queue management for asynchronous embedding.

Challenges we ran into

  • Managing vector indexes across different AI providers
  • Embedding consistency for candles vs. trades
  • Rate-limiting and batch processing with OpenAI and Vertex AI APIs
  • Scaling GitLab Runner and making sure deploys didn’t interrupt containers

Accomplishments that we're proud of

  • Successfully used GitLab to orchestrate an end-to-end MLOps pipeline
  • Designed reusable GitLab CI templates across services
  • Embedded over 20,000 candles and trades into MongoDB with consistent schema
  • Delivered vector search and GPT-powered query endpoints on schedule

What we learned

  • GitLab makes multi-repo workflows intuitive and highly customizable
  • You don’t need Kubernetes, Docker + GitLab Runner is enough for small MLOps
  • Semantic search changes how traders interact with their own data

What's next for NeuroTrade AI: GitLab-Driven Semantic Trade Intelligence

  • GitLab Issue Boards for user feedback integration
  • Auto-scaling services based on RabbitMQ queue load
  • GitLab Pages for public dashboards and doc hosting
  • Deeper model tuning with GitLab ML Experiments

Built With

Share this project:

Updates