Tinkoff Universe AI-assistants

The main articles are:

• Speech Recognition (Technology, Market)
• Speech technology: On the path from recognition to understanding
• RPA (Robotic process automation)

2024: Launch of the Assistant Universe

On April 24, 2024, Tinkoff announced the launch of the Universe of Assistants - AI assistants who know how to adapt to a person's requests and give him a personalized answer with recommendations or selection of goods on a variety of topics and issues - from travel planning to discussing new products in the film industry. The universe of assistants at the start is six AI assistants that differ from each other in special skills and interaction scenarios. Unlike other AI assistants who are trained to solve a wide range of issues for the user, each assistant of the Tinkoff Universe is trained to quickly solve problems of a deeper level in its field and process complex requests - for example, find sneakers with a certain color and pattern at a bargain price.

source = Tinkoff

As reported, the Universe of Assistants was created at the Tinkoff Center for Artificial Intelligence Technologies. Tinkoff assistants are based on specialized large language models (LLM), which are an alternative to ChatGPT.

Tinkoff assistants are personalized assistants who study every day. They can:

• adapt to the communication style of each user: from a child to a professional investor;
• when responding, consider the customer's past requests, preferences, and discussion context;
• work out atypical user requests using clarifying questions: from the search for a specific product to the plan of travel to a specific city;
• To support dialogue on any topic.

Based on the import substitution strategy, we have created our own technology and six assistants, each of which belongs to a specific direction of the Tinkoff ecosystem. They adapt to the user and adapt to their requests and preferences. noted Konstantin Markelov, Vice President, Business Technology Director at Tinkoff

source = Tinkoff

At the start, the Universe included the following AI assistants:

• Financial Assistant - will help you understand spending and give advice on how to better manage money;
• Shopping assistant - knows how to select goods from a specific category, taking into account individual needs (for example, understands the request: "Pick up a cream for dry hand skin in winter"), etc.;
• Travel assistant - will advise where it is better to go, pick up profitable flights and sign a travel plan;
• Invest assistant - will help you immerse yourself in the topic of investments and save time on studying the theory and analysis of securities;
• Junior assistant - knows how to entertain children and communicate with them on a wide range of topics, teach financial literacy and help to understand the school curriculum;
• Secretary - answers calls instead of a client, jokingly answers spammers, sends a record and transcript.

The universe of assistants turned out thanks to the work of 100 specialists and their worthy competence in the direction of LLM, as well as a sufficient amount of data and computing power that Tinkoff has. I am proud that our team knows how not only to work with large language models, but knows how to build products on their basis that really help users. And what is important, we do not make models that pass the exam, but we do only those models that solve the real problems of clients. told Viktor Tarnavsky, Director of Artificial Intelligence Tinkoff

At the start, users have access to the Invest Assistant in the Tinkoff Investments application.

In the near future, all assistants will be available to users in Tinkoff applications for IOS and Android, and some of them are still in Telegram. To use the assistant, the user needs to open a chat with him in Tinkoff mobile applications (Tinkoff, Tinkoff Investments, Tinkoff Junior, Shares), write a request, answer a couple of clarifying questions and receive a personalized answer with recommendations or product selections.

In the future, Tinkoff plans to expand and update the functionality of assistants based on user feedback and analysis of their needs, as well as add other AI assistants to the Universe.