Drop us a line
If you are interested in the development of a custom solution — send us the message and we'll schedule a talk about it.
Azati Software
Automated Steak Marbling Grading Solution
Our team developed an app that determines the marbling level of a steak from a photo and assigns it a quality grade. The quick analysis makes it easy to select a top-grade steak based on objective indicators.
Customer
The customer, one of the major players in the U.S. Agricultural sector, requested the development of a tool for standardizing and ensuring meat quality evaluation. This tool aims to guarantee high standards in steaks by grading marbling and categorizing steaks accordingly.
Objective
Our goal was to:
- Train a Convolutional Neural Network: (CNN) to recognize different steak images captured in production environments and accurately determine marbling levels. This involved optimizing the model to detect subtle marbling variations to assign the appropriate quality grade.
- Product Eligibility Rules Module: Manages and determines the rules for product eligibility and connection. This involves assessing customer profiles, current subscriptions, and applicable promotional rules to decide product offerings.
- Core Processing Module: Acts as the backbone of the system, ensuring smooth and efficient operation of all core functions. It handles essential tasks, system-wide operations, and the integration of various modules.
- Product Relationship Calculation Module: Calculates and manages relationships between different products. This includes handling dependencies, bundled offers, and interactions between various customer services.
Challenges
#1
CHALLENGE#1:
Problem: The client’s work schedule involved frequent business trips, leading to delays in responses and the provision of necessary information. This could slow down the development process and delay key project approvals.
Solution:
- We started by training the model using the dataset provided by the client.
- Simultaneously, we identified the must-have features for the mobile app and worked on them from a business analysis perspective, creating prototypes and detailed logic descriptions.
- We aligned all the critical aspects with the client at this stage.
- We shifted communication with the client to a written format, supporting all questions with prototypes and examples to ensure clear understanding and accelerate the approval process.
CHALLENGE#2:
Problem: Since steaks are perishable, they are stored in refrigeration units with bright lighting. During testing, it was found that the glare from the lamps on the meat was mistakenly identified by the AI model as fat streaks, leading to an overestimation of the marbling grade.
Solution:
- We discussed the issue with the client, and a solution was proposed.
- The client created a specialized “hood” for the device to block glare and prevent external lighting from impacting the accuracy of the analysis.
#2
Process
Pre-Sale Phase:
The first step involved calling the client to understand their problem and what they needed. Based on the discovery of market solutions and marbling assessment processes, we suggested how we could meet their needs.
Development Methodology:
The project followed the Kanban methodology, allowing flexibility and continuous improvement throughout the development process.
Development Methodology:
The project followed the Kanban methodology, allowing flexibility and continuous improvement throughout the development process.
Phase 1 – Model Training:
Initially, we used a dataset given by the client that contained steak pictures along with three evaluations from experts per photo to train the model. The model can accurately assess marbling levels.
Phase 2 – App Design and Logic:
We worked on designing the app’s screens and functionality, aligning everything with the client’s needs. Prototypes and detailed logic were shared for feedback and approval.
Phase 3 – Mobile App Development:
We developed the app as a minimum viable product (MVP) and then sent it for acceptance testing. The MVP allowed users to take photos and assess steak marbling.
Phase 4 – Post-MVP Iterations:
Based on user feedback, we refined the app in three additional iterations to enhance the meat quality evaluation process and user experience.
Solution
We developed and trained a CNN model that accurately evaluates steak marbling levels based on images. We integrated this model into a mobile application, allowing users to assess steak quality directly through their device’s camera.
And one of the important parts of our solution is that we can update model files without rebuilding the app. With this ability, it lets you upload any updated model file (like after the retraining step) directly to the app. This means that in the future users can be provided with an update on the model (in case of improvement or refinement) easily without having to download a new app version thus making it easier to maintain while also providing increased functionality.
Result
We successfully achieved all project goals, delivering a comprehensive solution for steak marbling assessment. Key accomplishments include:
- High-Accuracy Model: We trained the CNN model to accurately evaluate marbling levels from diverse steak images, ensuring consistent and objective grading standards.
- Cache Optimization: The process of loading product offerings into the cache was significantly optimized. We achieved a 15-20% improvement in cache “warm-up” times, which is a substantial gain given the large volumes of data involved.
- Flexible Model Updates: It is easy to update the model without rebuilding the entire app. This allows you to easily incorporate improvements and ensure high accuracy for your predictions for a long time.
- Enhanced Functionality Through Feedback: Through multiple post-MVP iterations, we fine-tuned and completed the functionality based on user feedback, creating a well-rounded tool tailored to real-world needs.
Overall, the project delivered an efficient, user-centered solution that is reliable, maintainable, and adaptable to future improvements.
Technologies
Featured case studies:
JavaScript, Ruby
HR Planning SoftwareThe customer asked Azati to audit the existing solution in terms of general performance to create a roadmap of future improvements. Our team also increased application performance and delivered several new features.
Python
Stock Market Trend Discovery with Machine LearningAt Azati Labs, our engineers developed an AI-powered prototype of a tool that can spot a stock market trend. Online trading applications may use this information to calculate the actual stock market price change.
Python
Semantic Search Engine for Bioinformatics CompanyAzati designed and developed a semantic search engine powered by machine learning. It extracts the actual meaning from the search query and looks for the most relevant results across huge scientific datasets.
Java, JavaScript
E-health Web Portal for International Software IntegratorAzati helped a well-known software integrator to eliminate legacy code, rebuild a complex web application, and fix the majority of mission-critical bugs.
JavaScript, Ruby
Custom Platform for Logistics and Goods TransportationAzati helped a European startup to create a custom logistics platform. It helps shippers to track goods in a real-time, as well as guarantees that the buyer will receive the product in a perfect condition.
