Smart store platform

kinect
kinect
kinect
kinect
kinect
telegram bot
telegram bot
telegram bot
quadcopter store simulator

Inspiration

Non-effective usage of online data in offline stores

What it does

We have two main directions: 1) Use online data to produce tools that can help in offline stores. To do that we use machine learning to clusterize goods and build guides and search engine based on the result. We are using telegram bots to show this guides. 2) Use data collected in store to boost sales. Without real store we can only build a tool to collect that data, so we created people tracking with Kinect, API to access tracking data and example that is using it. We also implemented AI for quadcopters that should use tracking data and help customers by using out guides.

How we built it

Guides and recommendations: We propose information retrieval on clusters using topic modeling (LDA, Latent Dirichlet Allocation). Then we introduce an entropy approach in guides generation. Each important word from the results of LDA modelling with cluster’s accuracy tuning is extracted and combined with the name or the value of product attribute. The result is a structured attribute_name – attribute_value description of each cluster (for example, there is a cluster with children mountain ski and also an another cluster with cheap men ski for beginners).

Challenges we ran into

Data sparsity, poor products information, language troubles.

Accomplishments that we're proud of

New approach to retrieve information about products from users behavior; Tracking people motion with API that provides a lot of opportunities.

What we learned

Efficiently work in team and extremely fast MVP provision.

What's next for Smart store platform

Thigs we can do without any help: Augment reality with extremely precise positioning by using our tracking API, automatic drones that can replace almost all salespersons, framework to set up all things quickly. Thigs that we can do only with support from some store: Customers matching with their online profiles, recomendations for all users by using machine learning with tracking systems, store reorganization tool with machine learning (similar to IKEA's labyrinths).