LLamasoft Supply Chain Blog

← Back to Blog

App Studio OpenX – Making Data Scientists’ Lives Easier One App at a Time 

By Clara Quigley  June 29, 2020

A few years ago, I was working as a data scientist consultant at a CPG company. My project was predicting the risk of service failures, or “cuts,” on promotional orders. I gathered the data from different systems and worked on my algorithm. This is the fun part of a project! Being creative and using analytics to address real pain points for my clients is my favorite part about being a data scientist. But a onetime analysis on my computer did not solve the problem for my client. To make the science drive business value, had to work for weeks to cobble together different parts of my workflow to “productionalize” my analysis. They needed to be able to interact with the outputs in a digestible form that would update automatically on a schedule. To do that, there are ETL processes that run in Azure Data Factory. I scheduled a script to run on a virtual machine and write data back to our database, and I had to schedule a different refresh of my dashboard after these steps were complete. Each of these processes had to be set up in different platforms with different languages and schedules. As a data scientist, I was never fully trained on productionalizing solutions – there isn’t a class for that in school! I learned a ton, but Im sure I wasted time testing approaches and learning new processes. I also had to lean on my software engineering colleagues in my IT environment to help with the backend details and user interfaceAnd in the end, the solution was painful to maintain and inefficient 

Even after putting in all that work to schedule and automate my scripts, the business users consuming my analysis could only interact with the final dashboard. As a resultI spent a lot of time answering questions about the input data (such as, “Are you using customer forecasts in your algorithm?”) or being asked to adjust parameters and provide additional output, outside the scheduled refresh. The business users never had the opportunity to own the analysis because they had no ability to interact with the endtoend process.  

The solution we really needed was a web application. (My colleague Aster Santana wrote a great essay on why enterprises need purposebuilt applications.) With an app, business users can log in and see the input data, run the model on demand, try a new scenario, and interact with the visualization layer. But as a data scientist, even with a master’s degree, building an enterprise-grade web application is not in my skillset.  

llama.ai App Studio OpenX 

This is where App Studio OpenX saves the dayAs part of the llama.ai platform, App Studio OpenX empowers data scientists like me to create productionready, scalable, and reusable applications without ever writing any code for the front-end interface or backend database. I can quickly share my work with the business users by simply sending a link. They can then log in, update data, change parameters, run the algorithm, and view the results and visuals – all within the web application.  

A typical App Studio OpenX application is composed of three parts: inputs, parameters and outputs.  

Inputs 

This is where users can upload input data, run scripts to pull data from API sources and interact with the data that will feed our application. We also enable visualizations to speed insight on the data’s completeness or quality. 

Parameters  

Here users can interact with the scripts running behind the scenes. Data scientists can set up parameters to incorporate user input into their algorithm. This could be as simple as allowing the user to set the date range for the analysis or as complex as setting which models should be run in a forecasting or routing optimization application. Then users can run the algorithm by simply pressing the play button.  

Outputs 

Here users can view custom visualizations in their favorite BI tool. (We like to use Tableau or Power BI.) They can also view the outputs in tabular form. Plus, users can set up scripts to export the results to the relevant system for the business problem, whether that is their ERP platform or emailing a report to a director.  

Why App Studio OpenX?  

App Studio OpenX allows data scientists to deliver commercialgrade applications quickly and in an agile fashion. Putting an app in your users hands early in the process provides crucial feedback and helps to incorporate the relevant domain knowledge into the application. This also increases adoption. When users can see the endtoend process, it empowers them to own the application and drive adoption across their teams – which increases business valueUsers don’t have to understand every nook and cranny of the algorithm to use the application and are empowered with artificial intelligence in a practical way. 

App Studio OpenX allows you to bring your own algorithm or use LLamasoft’s existing solver engineBy leveraging open source technologiessuch as Python and R, I am not constrained by which algorithms I can apply to solve the problem at hand. Any algorithm I can code, I can use (including calling App Studio OpenXs embedded Mathematical Commercial Solver). For example, in a project I worked on where we were trying to predict volatile near-term demand, my colleague Jason Silverman realized that a point prediction (or the mean) was not really that helpful. The amount of volatility meant that while the mean was right on average a , it wasn’t right in most weeks. What we really needed was a quantile prediction, or the probability that the order would be less than the value. For this, Jason used a random forest quantile regressor. This algorithm was perfect for the use case, but many standard packages and platforms do not provide algorithms that produce predictions at all quantiles. App Studio OpenX allowed us to bring the right algorithm to what has become a very powerful LLamasoft Rapid App. 

FinallyApp Studio OpenX empowers me as a data scientist to focus on what I like (and know) best – writing algorithms! I’m happier and more productive when I’m working on solving an interesting problem with an algorithm than when I am wrestling with databases and refresh schedules. 

Conclusion 

As companies mature in their AI journey, they will need more apps and fewer read-only reports. The best way to empower your own data scientists and get AI into the hands of your business users quickly is App Studio OpenX – a powerful part of the llama.ai platform!