Problem Description - 

Develop a ML system to determine relevant products for a store based on its past transaction history. These suggestions can be used in multiple sales channels to the store like self ordering app, sales executive PDA or by a calling agent to ensure that store is not missing out on any of these relevant products.

Train a model on the given dataset to predict each customer's next order

Participants may make the following assumptions

• Every store has max 1 transaction per day which may contain multiple products
• All stores served from the same fulfillment center are exposed to the same set of products. Hence, participants may create separate models per fulfillment center

Evaluation Criteria - 

• We will evaluate based on the recall score on test dataset
• Recall score will be calculated as - count of items which were predicted and bought / count of items which were bought
• Achieving a higher recall score indicates better performance

Submit your approach on how you will solve this problem. The full problem statement will be revealed if you are shortlisted for the next phase.

Problem Description

Given a store’s image, try to predict:

• Category of products store sells (bonus points if we can identify the exact brand and product as well)
• Whether the store is small, medium or large in size

Background

Computer vision can help better understand the store. Knowing the size of the store and the kind of products it sells would feed back into the personalisation engine for customized products and offers to the store. 

Stores exist in different sizes and shapes in India and hence it makes this image processing and analysis problem more interesting.

Participants may make the following assumptions

• Use the available information in the image to identify product categories. Broad categories we are looking for are Chips, Biscuits, Snacks (other than chips), Masala, Beverages, Cosmetics, Noodles, Soaps.
• Participants are free to add more categories as required. (e.g Bakery Items)
• Participants are free to drill down into categories. e.g. Beverages can be broken down into Beverages - Tea, Beverages - Coffee
• Exact dimensions of the store are not required. The mechanism of store classification needs to be explained in the approach
• Any appropriate annotation tool can be used
• A confidence score of 50% or above is acceptable, unless explained otherwise
• Participants need to provide precision, recall and MaP at various levels along with prediction outputs

Evaluation Criteria

• For category identification
• Percentage of categories accurately identified overall with confidence score of over 50% 
• New categories or sub-categories added and detected with confidence score of over 50%
• Precision, recall, MaP overall and per class
• Ability to detect in low-quality / blurred images
• Choice of algorithm
• For store identification
• Approach used for estimating store size
• Accuracy of classification based on precision, recall, MaP overall and per class

Submit your approach on how you will solve this problem. The full problem statement will be revealed if you are shortlisted for the next phase.

Floor design and SKU storage is a standard problem statement for any warehouse. For micro warehouses, a cost efficient and simple solution needs to be devised which can be used on laptop and mobile devices seamlessly, with below objectives.

A station manager of a warehouse should be able to upload a warehouse floor plan (2D) and sketch the storage areas per category (biscuits, soaps, shampoos, chips etc.) and tag those areas with category names.

• A simple floor mapping and tagging solution using javascript which can allow floor managers to custom mark storage areas for a warehouse floor and save the output in the form of a JSON document.
• Given the SKU dimensions, highlight the area's amount of SKU units which can be stacked over (you can assume a safe stacking height of say 3ft).
• Given an input invoice containing line items of SKUs and quantity, the utility should also be able to overlay a heatmap on the demarcated floor layout for the station manager highlighting different sized bubbles(based on qty of SKU in invoice) on demarcated areas.

A sample floor with tagged areas and highlighted heat map can be visualized as under -