Human machine interfaces

Case studies:

The problem

At the moment players can play video games against ‘ruled-based’ opponents not clever enough. There is the need to create agents (algorithms) able to play with/against players in a similar fashion of the ‘advanced chess’ today ( link text).

THE SOLUTION

Using reinforcement learning algorithms we have developed agents able to complete adventure games. The hybrid human-reinforcement learning algorithm is able to play the videogame 30% faster than any machine alone. This solution is then applied to drone control by instructions ('go to red door') instead of commands (left, right, etc.).

The problem

Voice messaging app aiming to allow the user to communicate with minimum interaction with the phone (or a walkie talkie replacement). Voice detection APIs currently available in the market are expensive not customized and legally binding (Alexa, Siri, Google Voice).

THE SOLUTION

We studied the recent research on sound detection and developed a voice detection API which is faster (20%) and more precise than the normal solutions available (40% more accurate). The app is now featured on medium and it is considered the best UK messaging app for people with disabilities (1 in 5).

The problem

Asset management managers and private equity analysts spend significant time just reviewing company reports, understand them and producing their own reports for their investors. The process at the moment is manual and time consuming.

THE SOLUTION

We build an interactive system (with a fine tuned LLM) that allows the manager to produce automatically quarterly reports (with sources' references on each paragraph) on each company of the portfolio just uploading the financial reports. The system allows Q&A on the documents uploaded and it's customised on financial information.

The problem

When a company wants to build an automated process (RPA based) from a manual process, it has to explain in details the current process to a consultant. Often, some information are not captured upfront (or given for granted) resulting in delays and more costs.

THE SOLUTION

We built an automatic system (with a multimodal model) to transform the manual process information directly into a robotic software development requirements. The solution automates the process of extracting and process information, it finds potential missing information and it plans and develop robotic solutions.

The problem

Judges are overwhelmed by lawsuits and certain types of cases are less subject to law interpretations (like car accidents and certain disputes). The goal is first to organise the documents to ease the search in the archive and then to test if an AI solution could deliver trustworthy sentences.

THE SOLUTION

First, we developed an (LLM-based) automated solution to labels court documents and organize automatically their search database (saved 1000s of hours in processes and simplify access of the citizen data). We then designed a POC to automatically judge car accidents lawsuits giving a reasoning with each sentence. We reached an accuracy above 95%.

The problem

Brain computer interface is a new technology that enables to give insights on human thoughts and feelings based artificial neural network analysis of brainwaves. Few institutions and companies now have the expertise in these area.

THE SOLUTION

We are currently developing some new technology based on deep learning models to understand feelings and thoughts in specific scenarios.

How can we help you?

Email: contact@aitechnologies.co

Copyright @ AI Technologies 2022 All Right Reserved.

Privacy Policy