Is AI – ML – DL Applicable?
We believe that in near future Artificial Intelligence will be the top priority requirement in any kind of production software no matter the vertical industry.
Everything can be designed; few things are designed better!
BitRezus has always revolved AI around Blockchain as a result of the core team’s expertise in artificial intelligence, data science, statistical modelling, visualizations and big data predictive analytics. We are assisting businesses to innovate in solving their real-life business problems by working with them to solve challenges using exploratory data analysis, feature engineering and selection, model building with fit-for-purpose algorithms, model training & validation, model evaluation and providing easy-to-understand results and visualizations.
We also help enterprises and startups in designing, installing, operating and optimizing their AI/ML pipeline into your automated CI/CD pipeline. We have a team of professional Senior Software Engineers, Enterprise Architects and Data Scientists that will take your “sandboxed” research environment to end-to-end production-grade, performant, scalable and reliable system.
BitRezus scopes down each ML project into the following phases with their respective time estimation:
Phase I: ML Project Plan Setup (2-4 weeks)
ML project plan starts with requirements gathering and evaluate whether machine learning meets the business goals. Next, we try to understand what kind of data is available, how the data is acquired, sources of data and verify the quality.
In this stage and depending on the project requirements, we try to specify the type of algorithms to be used (supervised/unsupervised/semi-supervised), in order to be able to fully describe the scope and goal of the Proof-of-Concept (PoC) along with its accuracy.
Finally, we discuss the PoC together with the client and agree whether it brings substantial value to its services and/or products and is worth investing.
Phase II: Model Exploration (4 - 8 weeks)
In this stage, our engineers actually build the PoC model, train and evaluate the baseline model using data from Phase I. After the baseline model is built and trained, we proceed with a goal performance estimation for the production-model.
At the end of this stage, we should be able to provide with a more realistic estimation of the performance of the production model, given the metrics from Phase I.
Phase III: Development (3+ months)
In this phase, our team works iteratively in building the production model and trying to reach the agreed performance/accuracy goals. That may sometimes mean that our team may need to change or tweak something in the algorithm to perform better, try out a different method or even adjust the data, until the goal is reached.
During development of the production model, our team works in sprints of 2-3 week duration and we engage with the customer along the way, to get his feedback as early as possible. The outcome of this phase is the first production-grade model which the customer can use.
Phase IV: Model Improvement (indefinite duration)
After model deployment into production, it is vital that the model is being monitored and evaluated at certain time intervals, to make sure its prediction capacity does not degrade as new data patterns come into play and that the initial performance and business goals are met.
Machine learning models take time to show substantial results so in BitRezus we believe that this phase is crucial for the ML project to be successful.
Final Remark: Model Deployment into Production (2-4 weeks)
Our engineers help you along the way to successfully integrate the AI/ML pipeline into your continuous integration/continuous development environment by maximizing model reproducibility, compatibility, generality, and scalability in essentially every AI/ML stage discussed previously. In this stage we strive to make sure that the predictions obtained during prototyping and research are identical to the predictions acquired in the production environment. In that way, we can assure that we have a reproducible pipeline which is optimized, with minimum cost and effort for the enterprise and the team.