We are specialising in developing tailored, bespoke Machine Learning solutions for various businesses. In our products, we also incorporate state of the art AI components like Google Cloud Vision or Google Text To Speech.

Our Machine Learning know-how can be utilised in all sorts of domains - from creative-tech to gaming, messaging, human-machine collaborative tools, automatic problem solving, data insights computation as well as data acquisition, augmentation and processing.

Chatbots are a great way of automating communication. They can be text- or voice-based and they can live across pretty much any messaging platform. At Le Polish Bureau, we have experience in delivering solutions for:

• Slack
• Facebook Messenger
• Google Assistant
• Amazon Alexa
• Web

Our natural language processing solutions are based on Google Dialogflow, providing highest level of semantic analysis, combined with support of multiple languages and deployment platforms.

A great example is OnePlus: Crackables, where a chatbot we developed was driving an online puzzle contest.

Sometimes you might have a specific need or an idea that you don't necessarily have all the needed data for. And gathering the data by your team will likely take ages. Probably literally.

This is where our Machine Learning data crowdsourcing solutions come to play.

We implemented this approach in order to analyze people's perception of the phenomenon of the northern lights. We developed a custom interactive questionnaire that allowed respondents to group the auroras by visual similarity. A different questionnaire was then asked to match audio to visuals in order to train the AI about "what an aurora could sound like".

There doesn't always have to be a clear scientific task to be solved with Machine Learning though. AI can be used for creative purposes too.

One of these is Neural Style Transfer. A use of AI to generate completely new visuals or audio based on reference style and a target piece to stylise.

Another interesting domain is reinforcement learning. It relies on the ability to run tens of thousands of parallel, sped-up simulations of a usually mechanical task that the AI is supposed to learn. For example, throwing balls to a basked. Without having to be taught explicitly how to do this, an AI can master such task quickly:

Another domain we specialise in is Machine Vision. We are using state-of-the-art implementations from Google to analyse the world around you. We can also implement custom Artificial Neural Networks that are specifically trained to perform a new detection task.

See how we implemented Machine Learning solutions before: