Research & Development
SMART BAKE AI
Scientific and technological research aimed at removing the scientific and technological uncertainty in the modelling and optimisation of dough-based bakery product recipes through the development of experimental artificial intelligence models.
Subject and objectives
What the project investigated
The subject of the project was the systematic scientific investigation of the relationships between recipe composition, raw materials, production conditions and the final quality characteristics of dough-based bakery products.
Before the project, the development of new or modified recipes relied mainly on experience, on repeated trial production runs and on the gradual adjustment of quantities and conditions. The process involved increased uncertainty, required a significant number of experimental trials and did not ensure that conclusions drawn from one recipe could be reliably applied to different categories of dough-based bakery products.
The main objectives
- Scientific investigation of the relationships between raw materials, dosages, production conditions and quality characteristics.
- Identification of the critical parameters that affect the behaviour and quality of dough-based bakery products.
- Organisation and pre-processing of the experimental data.
- Development, training and comparative evaluation of different artificial intelligence and machine learning models.
- Development of experimental models for predicting the quality characteristics.
- Investigation of the scope for optimising recipes and raw material dosages.
- Development of an experimental web-based software version for entering selected parameters and producing experimental predictions.
- Evaluation of the models in terms of accuracy, stability, reliability, generalisability and interpretability.
- Recording of deviations, unsuccessful approaches and technical limitations.
- Production and documentation of new scientific and technological knowledge.
The innovation of the project lies in combining organised experimental production data with algorithmic methods of artificial intelligence and machine learning, with the aim of developing experimental models for prediction and optimisation. The simple application of an off-the-shelf algorithm was not sufficient.
Work packages
WP1 — WP4
The project was organised into four functionally interconnected work packages, with a clear allocation of responsibilities, distinct time milestones and a review of every deliverable.
WP1
Production of experimental test specimens of dough-based bakery products and collection of research data
Implementation period: 01.01.2025 – 31.12.2025
WP1 formed the experimental basis of the project and was carried out throughout the 2025 financial year. Its subject was the systematic production of test specimens and experimental batches and the creation of an organised set of real experimental data. Each experimental batch was treated as a distinct research unit.
For each experimental trial
- Selection and procurement of raw materials and consumables
- Weighing, dosing and preparation of ingredients
- Preparation of test specimens and experimental batches
- Application of planned variations in composition and dosage
- Monitoring of temperatures, times and production conditions
- Monitoring of storage conditions
- Evaluation of thermal behaviour, spoilage, shelf life and overall quality
- Recording of deviations, failures and unsuccessful approaches
- Classification and organisation of the data and measurements
Preliminary scientific investigation
- Study of international scientific literature and technological trends
- Analysis of the biochemical and physical mechanisms of fermentation
- Investigation of the parameters affecting taste, texture and volume
- Recording of flour types, raw materials and special-diet additives
- Analysis of heat treatment parameters
- Definition of the requirements for recipe optimisation through artificial intelligence
- Investigation of existing technological limitations in the baking industry
- Design of the system architecture
- Definition of quality indicators (texture, volume, moisture, sensory indicators)
WP2
Scientific investigation, analysis of experimental data and initial technological design
Implementation period: 19.11.2025 – 30.11.2025 · Deliverable D1
The first work package of the subcontractor’s specialised scientific and technological work, based on the data from WP1. The available data were received, recorded and classified, and an assessment was made of the completeness of the records, the consistency of the nomenclature and units of measurement, the presence of missing values, the suitability for algorithmic processing and the matching of input parameters with results.
Particular emphasis was placed on identifying the scientific and technological uncertainties: it was examined whether the available data could capture with sufficient accuracy the complex relationships between composition and final quality, and whether it was feasible to develop models with acceptable stability and generalisability.
On this basis
- The research hypotheses were formulated and the research objectives were defined
- The data pre-processing procedure was designed
- The initial architecture of the algorithmic models was determined
- Data processing and the training of initial models began
- The first version of an algorithmic model was developed and experimentally evaluated
Full documentation in Deliverable D1 →
WP3
Development and evaluation of an experimental artificial intelligence software version
Implementation period: 01.12.2025 – 31.12.2025 · Deliverable D2
The core work package of experimental software development. The methodology comprised the cleaning, normalisation, transformation and organisation of the data, followed by the development of input features that captured the relationships between composition, production conditions and quality characteristics.
Different algorithmic approaches were investigated comparatively. For each approach, cycles of initial parameterisation, training, experimental evaluation, recording of deviations and limitations, retraining and parameter optimisation were carried out.
The selected models were integrated into an experimental web application. The implementation was not a finished commercial product, but experimental software for evaluating the technological feasibility of the models developed.
The experimental application and its documentation →
WP4
Final evaluation, documentation and publication of the research results
Implementation period: 01.01.2026 – 31.01.2026 · Deliverable D3
The continuation and completion of the project. Its subject was the collection, synthesis and final evaluation of the results obtained from the experimental production, the data processing, the training of the models and the operation of the experimental application.
- Collection and comparison of the results of the algorithmic approaches
- Assessment of the extent to which the initial uncertainties were reduced
- Recording of the final evaluation metrics
- Documentation of deviations and unsuccessful approaches
- Recording of technical limitations and open research questions
- Formulation of the final conclusions and documentation of the new knowledge
Timeline
01.01.2025 — 31.01.2026
Total duration of 13 months.
| Work package / Deliverable | Jan 25 | Feb | Mar | Apr | May | Jun | Jul | Aug | Sep | Oct | Nov | Dec | Jan 26 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| WP1 — Production of experimental test specimens and data collection | ■ | ■ | ■ | ■ | ■ | ■ | ■ | ■ | ■ | ■ | ■ | ■ | |
| WP2 — Scientific investigation and initial design | ■ | ||||||||||||
| D1 — Online documentation of the methodology | ● | ||||||||||||
| WP3 — Development and evaluation of experimental software | ■ | ||||||||||||
| D2 — Experimental AI software version | ● | ||||||||||||
| WP4 — Final evaluation, documentation and publication | ■ | ||||||||||||
| D3 — Online final evaluation report | ● |
Review milestones
30.11.2025
Completion of WP2 and delivery of D1
31.12.2025
Completion of WP3 and delivery of D2
31.01.2026
Completion of WP4 and delivery of D3
Each deliverable was reviewed in terms of its completeness, its consistency with the contractual subject, the functionality of the electronic link and the adequacy of the technical and scientific documentation.
Technological tools
Two complementary levels of infrastructure
Experimental production infrastructure
The professional premises, fixed assets, equipment, raw materials and consumables used for the production of the experimental test specimens and the collection of the WP1 data, under repeatable and, as far as possible, controlled conditions of temperature and humidity.
Computing and web infrastructure
Environments for data processing and organisation, for the development and evaluation of algorithmic models, as well as a web environment for hosting the documentation and the experimental application.
What the computational methodology supported
Deliverables
Three online sections
D1
Online documentation of the research methodology and the initial experimental results
Deliverable of WP2.
30.11.2025 smartbakeai.com/research/documentation
D2
Experimental artificial intelligence software version for the modelling and optimisation of dough-based bakery product recipes
Deliverable of WP3.
31.12.2025 smartbakeai.com/research/software
D3
Online final evaluation report and results of the scientific and technological research
Deliverable of WP4.
31.01.2026 smartbakeai.com/research/results