TÜBİTAK-Supported R&D at Cotcast: Strengthening Our Forecasting Engine

How our forecasting engine works and what this R&D strengthens

Cotcast’s forecasting engine is built on a multi layer data and model architecture, rather than relying on a single data source or a single model. Price data, supply demand indicators, weather and climate signals, macroeconomic variables, and market sentiment related data layers are combined within one unified flow. This allows the system to go beyond the question “where will price go” and also address “what do the underlying conditions behind this move indicate.”

With this R&D effort, our goal is to strengthen the signals produced by the engine in three ways. First, we are building a validation and testing framework that allows us to measure signal reliability more clearly. Second, we are optimizing the pipeline from data to signal so we can generate cleaner outputs in less time. Third, we are making these outputs more visible and understandable inside the product, so they serve decision making more directly.

1) Stronger validation and testing infrastructure

In forecasting systems, the key differentiator is not only producing “good forecasts,” but being able to measure how reliable those forecasts are under different conditions. For this reason, we treat validation as a product discipline within our R&D work.

We are building a framework that can evaluate performance across different market regimes, volatility environments, and trend structures. Instead of relying on a single average metric, this helps us understand when the engine performs best and why. We compare new improvements against existing versions under the same conditions, and we only ship changes that produce measurable gains.

This approach provides a strong trust signal for both investors and business partners. As the system evolves, the measurement standards mature with it.

2) Faster signal generation and a more efficient analysis flow

In commodity markets, the value of a decision is often defined by timing. That is why the flow from data to insight must be fast, repeatable, and clean. As part of this R&D work, we are optimizing the data processing pipeline and the signal generation steps to make the overall workflow more efficient.

Our goal is not only speed. We also aim to make the signal generation process simpler and more traceable. This improves how signals are presented in the product and makes them easier to interpret. As a result, users can follow market moves earlier and connect signals to their decisions more quickly.

3) Outputs that directly translate into product value

If R&D outputs remain only on the technical side, user value stays limited. That is why we treat product integration as part of the design from day one. Our goal is to transform the signals produced by the engine into more actionable decision support within the product.

This approach delivers two outcomes. First, the information presented to users becomes more actionable. Second, the product develops a consistent language of presentation. Signals are understood not only as outcomes, but also with their context. This is one of the core drivers of long term user trust.

From cotton to soybean: how we transfer our expertise to new commodities

Cotcast’s initial focus area was cotton. The data layers, modeling practices, and market specific expertise built in cotton provide a strong foundation for expanding into other agricultural commodities. Through this R&D effort, we are scaling that foundation so it can be adapted to new products such as soybean.

Cotton and soybean have different market dynamics, but they share a common ground. Both are sensitive to weather conditions, global supply demand balance, and geopolitical risk. For this reason, our approach is not “copy the same model to every commodity.” We keep the same architecture, then adapt the model with commodity specific data sources and parameter sets aligned with the behavior of each market.

On the soybean side, we incorporate reports and market indicators that reflect the nature of the product. This enables the same infrastructure to support a consistent forecasting and signal generation setup across multiple commodities.

Scalability: infrastructure for new commodities and growing data volume

New commodities mean more data sources and more users. For this reason, strengthening the backend with a modular and scalable structure has been a critical focus of this R&D work. Bringing data layers together within a single pipeline makes modeling more consistent while also reducing operational complexity.

In short, our objective is clear. As data volume and the number of supported commodities grow, system performance should remain stable. When a new commodity is added, it should be deployed quickly while preserving the same validation and testing standards. This supports Cotcast’s long term vision of becoming a multi commodity decision support platform.

What we will share as the work progresses

As this R&D effort advances, we plan to share product facing updates through key milestones. This will help us communicate both technical progress and product value more transparently. Our priority is user impact and measurable improvements.

cotcast.ai