Research · Deliverable D3
Online final evaluation and results report of the scientific and technological research
| Work package | WP4 |
|---|---|
| Delivery date | 31.01.2026 |
| Status | Completed |
| Web link | smartbakeai.com/research/results |
D3 is the final deliverable of the project and comprises the synthesis and final evaluation of the experimental results, the assessment of the degree to which the initial uncertainties were reduced, and the documentation of the new knowledge produced. This is a translation. The authoritative version of this deliverable is the Greek one.
1
Overall outcome of the project
The project led to the production of new scientific and technological knowledge regarding the possibility of systematically correlating recipe composition and production conditions with the final quality characteristics of dough-based bakery products, and to the experimental investigation of the possibility of using artificial intelligence techniques to predict and optimise them.
The critical point. The research was not limited to applying off-the-shelf artificial intelligence methods, but included iterative development, testing, comparison, rejection or improvement of different approaches. The result was the acquisition of documented knowledge about both the capabilities and the limits of using artificial intelligence in the modelling of the specific experimental data.
2
Assessment against the initial objectives
| Initial objective | Outcome |
|---|---|
| Systematic investigation of the relationship between raw materials, dosages and production conditions and quality | An organised set of experimental data was created from repeated trials, and the critical input variables and the results were recorded |
| Identification of critical parameters | The effect of composition, dosage, temperatures, times and production and storage conditions was investigated |
| Development of artificial intelligence models | Different algorithmic approaches were developed, trained, retrained and experimentally evaluated |
| Development of prediction and optimisation models | Experimental models were implemented to investigate the prediction of quality characteristics and the optimisation of recipes |
| Comparative evaluation of different approaches | Iterative evaluation cycles were carried out and the models were compared in terms of accuracy, stability, reliability, generalisation and interpretability |
| Development of experimental software | A functional experimental version of artificial intelligence software was developed and published online |
| Reduction of scientific and technological uncertainty | Documented knowledge emerged regarding both the possibilities and the limitations of applying the models to the specific experimental data |
| Production of new knowledge | The successful and unsuccessful approaches, the technical limitations and the conclusions on the modelling and optimisation of the recipes were documented |
The assessment demonstrates that the main research and technological objectives of the project were achieved.
3
Evaluation metrics
Within D3, the overall assessment of the experimental results and of the metrics of accuracy, stability and generalisability was carried out, together with the recording of the deviations, the unsuccessful approaches, the technical limitations and the degree to which the initial scientific and technological uncertainties were reduced.
MAINTENANCE NOTE — must be completed before publication.
This is where the actual numerical metrics from the subcontractor’s technical documentation must go: a table per algorithmic approach with error metrics, results on data outside training, stability indicators and the decision to accept or reject.
These values are not estimated and are not filled in approximately. This is the point that is checked first in any audit of the project file.
4
Technical limitations and open research questions
Recording the limits of the technology developed is an essential result of the project and distinguishes research from the simple application of known techniques.
Range of validity of the models
The predictions hold within the experimental range of the training data. Outside it, this is extrapolation with significantly reduced reliability.
Raw material variability
The models do not capture the quality variability of raw materials per supplier batch, nor the variability of the production environment.
Generalisability
Generalisation to categories of dough-based bakery products outside the training set remains an open research question.
Sensory indicators
These are indirect estimates and do not replace organoleptic evaluation.
5
Economic and commercial exploitation
The exploitation of the results is primarily internal to the business. The new knowledge, the organised base of experimental data and the models developed create the conditions for the business to support well-documented decisions when designing and modifying recipes, to reduce dependence on purely empirical procedures and to direct subsequent experimental trials more effectively.
In future, this exploitation may contribute to reducing the number of unproductive trials, the consumption of raw materials and the time required to develop new products. After further technological maturation and validation, the technological solution may also be used in a wider professional environment for the production of dough-based bakery products.
Target groups
Direct target group
The business itself, the research team and its external scientific collaborators, who acquired new know-how regarding the organisation of experimental data and the development of artificial intelligence models.
Wider potential groups
Bakery businesses and producers of dough-based bakery products, producers of functional or special-diet products, food processing businesses and research or technical teams seeking to make use of computational methods.
6
Dissemination and publicity
The dissemination of the results was carried out through a purpose-built online research environment. The methodology, the experimental development and the results were organised into three distinct and functionally accessible sections.
D1
Research methodology and initial experimental results
30.11.2025 smartbakeai.com/research/documentation
D2
Experimental version of artificial intelligence software
31.12.2025 smartbakeai.com/research/software
In this way, the organised electronic presentation of the research methodology, the experimental technological development and the final results was ensured, while at the same time the ability to protect individual data and know-how that are not intended for public disclosure is maintained.