Data is available to almost all small to medium scale businesses now-a-days. All shops have some sort of an electronic data collection and saving system be it in a printed or csv format. But these businesses cannot leverage this data to scale up or make better stocking decisions and such. Our product, DataMagic, allows anyone without any kind of understanding of Data Science and statistics to use data generated by them to predict things that matter for their business. We have developed an intuitive web interface that allows the user to upload their data in cvs or jpg(ocr will be done in a later work). Then they are instructed to select the feature that they want predicted(like quantity, expenditures etc.) and select the columns which houses data like time or date and unique products for which the features have to be predicted. This data is passed on to our backend which preprocesses the data automatically and makes it suitable for time-interval learning. We then use a Recurrent Neural Network to train a model on the data. This model is further used to predict the quantity or expenditure of upcoming days/weeks/months and the predictions are visualised on a graph along with the currently available real data.
AutoML is a huge problem to tackle specially when you wish to generalise the entire process in 24 hours. We had to rethink immense amounts of fundamental details of our idea throughout the hack. We decided to stick to just small/medium businesses choosing to not consider senarios like educational institutions and such. We have taken care to keep the web-interface very intuitive and not throw around terms like 'labels', 'features' etc.. to confuse the user. After the user uploads the data, instead of giving them the predictions in a numerical manner, we plot them with the existing data on charts. Linking the charts and passing values from the python backend proved to be a task. We also had to write optimised code for training the model so that the processing can happen in real time and not make the user wait.