In the fast-paced world of automated trading, where computational efficiency meets market intelligence, selecting the right tools can transform a novel idea into a competitive advantage. For Maxvankekeren-IT, a Dutch technology company that evolved from web hosting to AI-powered trading, the implementation of AgileRL marked a decisive turning point in their ability to compete in the financial markets while maintaining efficient resource utilisation.

"AgileRL has massively increased our productivity in developing, training and experimenting with new models. It's super easy to use and is resource efficient." — Max van Kekeren

‍

From Web Infrastructure to AI Innovation

Maxvankekeren-IT's journey represents a fascinating evolution in the technology sector. Originally established as a web hosting company, they built their foundation on providing reliable application and website hosting services.Under the leadership of Max van Kekeren, who brings a strong background in artificial intelligence, the company developed sophisticated internal AI systems for various applications:

• Automatic cyberattack detection
• SPAM filtering
• NSFW content filtering
• Server resource usage prediction

This expertise in AI applications led to an innovative venture: developing AI-powered trading bots initially designed to grow the company's emergency savings fund. What began as an internal project has evolved into a specialized service, where they partner with investors through a Dutch exchange, managing investments using their sophisticated AI trading systems while operating on a profit-sharing model.

‍

Breaking Free from Traditional Methods

Prior to implementing AgileRL, Maxvankekeren-IT's trading approach relied on a combination of time-series forecasting models and algorithmic trading. While functional, this solution revealed several critical limitations:

• ‍Market Adaptability: Traditional algorithmic trading struggled to maintain flexibility and creativity in challenging market conditions, often missing opportunities that required more sophisticated decision-making.‍
• Resource Constraints: Their initial venture into reinforcement learning using Stable-Baselines3, while promising, proved extremely resource-intensive, constantly pushing their servers to 100% CPU and GPU utilisation.‍
• Development Speed: The inefficient training process with Stable-Baselines3 created a significant bottleneck in their ability to prototype and iterate on new trading strategies.

These challenges were particularly significant for a small team aiming to democratize advanced trading capabilities while operating with limited computational resources.

‍

Discovering and Deploying AgileRL

The decision to start working with AgileRL approximately 1.5 years ago marked the beginning of a transformative period in Maxvankekeren-IT's development capabilities. The process started with framework selection: AgileRL's implementation of RainbowDQN provided the sophisticated decision-making capabilities they needed, combined with remarkable efficiency in resource utilisation.

The team followed a methodical approach to implementation:

1. Initial prototyping using documentation examples
2. Development of custom trading environments
3. Parallel operation with existing systems for paper trading
4. Gradual transition through monthly updates

The team developed sophisticated integrations to enhance their trading capabilities:

• Real-time integration with crypto/stock exchanges for market data
• Implementation of Google Trends data for market sentiment analysis
• Development of LLM-based systems for social media sentiment analysis

‍

Quantifiable Improvements

The impact of implementing AgileRL on Maxvankekeren-IT's trading operations was substantial and measurable:

• ‍Development Throughput: The most dramatic improvement came in their ability to prototype and test new strategies. The team increased their concurrent model testing capacity from 2 to 16 experimental models, an eightfold improvement in development capability.‍
• Training Efficiency: Overall training performance improved by approximately 20%, while simultaneously reducing resource utilisation, allowing them to train larger and more complex models within their existing infrastructure.‍
• Resource Optimization: Where previous implementations frequently maxed out server resources, AgileRL's efficient processing allowed them to train more sophisticated models while maintaining comfortable resource headroom

‍

8x ‍Concurrent model‍testing capacity

+20% ‍Training performance

‍

This implementation demonstrates how AgileRL can enable a small, innovative team to compete effectively in the sophisticated world of automated trading. Maxvankekeren-IT's journey from web hosting to AI-powered trading showcases how the right framework can democratize advanced trading capabilities, making institutional-grade trading technology accessible to organizations operating with limited computational resources.
‍

If you would like to take part in a case-study with AgileRL, please reach out on LinkedIn.