FxChartAI ABT: The Final Piece on AI-Driven Forex Analysis Journey

Abiodun Aremu
4 min readNov 18, 2024

--

Over the past two months, I’ve had the privilege of introducing FxChartAI to users worldwide — a product born from training a transformer model on Forex data. Throughout this journey, I’ve chronicled insights in previous articles (links provided below) and shared real-world results on my social media platforms. Now, I’m excited to present the final chapter in this story.

https://abiodunaremung.medium.com/lessons-learned-from-fine-tuning-a-transformer-model-on-forex-data-d1d25e98bba0

The Journey to FxChartAI

Developing FxChartAI has been an exciting and rewarding experience. As mentioned in my earlier articles, I deployed the model on Replicate and built a web application that enabled real-time access to the model’s inferences — all within a week. Following this, I launched FxChartAI on ProductHunt, where the overwhelmingly positive feedback validated its potential value for both institutional and retail traders.

Understanding that no initial product is perfect, I prioritized collecting user feedback to refine FxChartAI. The users — primarily Forex traders seeking reliable market insights — relayed the challenges they faced in analyzing market trends and processing overwhelming amounts of information. FxChartAI simplifies this process by projecting market trends and offering traders actionable insights before they occur, reducing analysis time and minimizing information overload.

Iterating for User Needs

A week after launch, user feedback highlighted the need for insights across multiple timeframes, particularly higher ones. Initially, this posed a challenge since the model was trained on a 10-minute (M10) timeframe. However, with some fine-tuning, the model began performing well on higher timeframes, including M30, H1, H2, H8, D1, and WK. After further testing, we decided to limit availability to M10, H1, and H2, which struck a balance between precision and user accessibility. This enhancement significantly boosted user excitement and engagement.

Subsequent meetings revealed a need for more clarity in projected candle insights. Users found the 30 projected candles excessive and overwhelming. Addressing this, I refined the dashboard to project 12 candles, alongside detailed insights such as:
- Percentage confidence on the next candle’s position,
- Expected trend direction,
- Current candle position, and
- Expected bullish and bearish spikes.

These updates dramatically enhanced the product’s utility, resulting in more profitable trading decisions.

Trading result using FxChartAI on a demo account

Refining the Dashboard

Building on this momentum, I added the ability to view signals across all timeframes simultaneously, enabling users to compare projections side by side. Additionally, I integrated the full signal history of the M10 timeframe — covering the last 14 projections — directly into the dashboard. These changes empowered traders with a high-level overview of market conditions at a glance.

With these features in place, FxChartAI delivered more intuitive signals, bringing us closer to an optimal user experience.

FxChartAI’s new dashboard

Automated Back-Testing (ABT)

The next step was conducting Automated Back-Testing (ABT) on the model’s projections over a two-month period. This process had to wait until critical APIs for signal generation were integrated. The ABT results were encouraging, underscoring the model’s effectiveness while revealing areas for further optimization. The success of this experimental version validates that, with quality data and refined techniques, FxChartAI has immense potential for precision and reliability.

FxChartAI Conservative Automated Back-testing Result

FxChartAI and the Padroverse Vision

FxChartAI is more than a standalone project; it’s a proof of concept (PoC) for Padroverse, a more ambitious initiative. Padroverse seeks to create digital twins of physical entities, providing insights into anomalies and opportunities for optimization in the physical world. As I begin preliminary work on Padroverse, I’m excited to share updates on its progress in the future.

For now, FxChartAI has been withdrawn from commercial availability and operates as an internal product. This decision mitigates regulatory risks and allows us to focus on research and iterative improvements in collaboration with traders who can provide immediate performance feedback. Given the highly regulated nature of the Forex market, this approach ensures responsible development while safeguarding against misuse.

Signing Off

This article marks the conclusion of my journey with FxChartAI. It’s been a fulfilling experience, combining AI innovation with practical applications in Forex trading. I hope these insights have inspired you — whether you’re a trader, investor, AI product entrepreneur, or engineer.

Feel free to connect with me for discussions, collaborations, or questions:
- Instagram/Twitter/X: [@abiodunaremung](https://twitter.com/abiodunaremung)
- LinkedIn: [Abiodun Aremu] https://www.linkedin.com/in/abiodun-aremu-0756ba90
- Email: abiodunaremung@gmail.com

Thank you for joining me on this journey. As I embark on Padroverse, I look forward to exploring new frontiers in AI innovation. Stay tuned for what’s next, and I’ll see you at the next milestone.

Happy trading and innovating!

--

--

No responses yet