AI-driven question-answering system for learners

The solution provided round-the-clock student support, enhanced engagement rate, and freed teachers to focus on high-level instruction.

Data science & AI EdTech

USA

Education

Since 2024

Key achievements

1	month project duration
>1000	requests per minute system capacity
5+	points NPS growth

Challenge

The educational platform is used by students of all ages, from Kindergarten to high school. The platform already had a chatbot, but its current functionality did not meet the provider’s growing needs.

Teachers were not always available to answer all questions and answer students fully, especially across multiple domains.
Due to the high competition in the market, the customer was looking at new ideas to stay on top. They decided to use AI to facilitate decision-making, improve learning culture and adopt new forms of learning.
The main prerequisite for the AI-driven question-answering system for learners was the highest degree of data protection both within the system and towards any third parties.

Solution

After a series of joint meetups, it was decided to utilize the existing platform’s functionality and integrate the new AI-based tool with it. The chatbot should:

1
Conduct basic administrative tasks (e.g., replying to student’s questions in various domains)
2
Preserve the request history and maintain context in the dialogue process
3
Process an ever-growing variety of educational materials in the system
4
Provide personalized learning support depending on the pupil’s age
5
Integrate with the existing systems seamlessly
6
Ensure low query latency to keep students engaged
7
Have a high level of security and ensure no PII leakage
8
Handle multiple student inquiries simultaneously

Project scope

For the project, the Aristek team relied on subject matter expertise in AI-based development and the education domain. It consisted of the following steps:

Pipeline for data collection and processing.
Data extraction and embedding calculation.
Setting up a question-answering system utilizing retrieval-augmented generation techniques.
System testing (load up to 1000 requests per minute).

How it works

The pupil asks a question about the water cycle via the learning platform’s chat interface.

The AI chatbot processes the query, using its knowledge base and context from previous interactions to formulate a comprehensive answer.

The platform displays the response, along with options to explore related topics or ask follow-up questions.

The pupil can then delve deeper into the topic or explore suggested related concepts.

If needed, the chatbot uses adaptive questioning to assess understanding, refining its explanations until the student’s query is fully addressed. Depending on the student’s grade, the answer is presented differently.

Screenshots

Key features

AI systems such as ChatGPT are highly criticized as they may create text that seems credible at first glance but may in fact contain errors or nonsense. Thus, the trustworthiness of the answers is always questioned.

Taking this fact into account, the question-answering system was implemented with the ability to retrieve an excerpt from the training material and a link to its location.

Such tooltips are shown as in-app messages. Their key features include:

1
Providing contextual excerpts
When a user asks a question, the AI system retrieves a relevant excerpt from the existing database of training materials. Instead of generating a full response, the tooltip displays a concise, pre-existing text chunk that directly answers or clarifies the query. This ensures accuracy and helps avoid incomplete or incorrect answers.

How it works: If someone asks about “Newton’s laws of motion,” the tooltip could provide a short text from the course material that explains the first law, such as: “Newton’s First Law states that an object in motion will stay in motion unless acted upon by an external force.”
2
Streamlining access to further information
It can be cumbersome to find what you need instantly – it can take hours. For our solution, it takes an average of three milliseconds to provide an answer. Tooltips augment excerpts with a link to the full location of the content within the learning material. This saves a lot of time searching for the right paragraph or article in the institution’s database.
3
Increasing adaptability to any tone of voice
There’s no need for the user to enter the precise query. The system supports human-like communication and provides the answer according to the search meaning, ensuring that even vague or incomplete questions are met with relevant responses.

Tools & technologies

Python

GitHub

GPT-4

Jira

OpenAI

Team

Data scientist x2

Results

An AI-driven question-answering system for learners has revolutionized a wide range of application areas simultaneously.

The project is still in the metrics analytics phase, but the following improvements can already be noted:

Student questions are resolved in seconds, and answers cater to previous student inquiries
Ability to provide a vast variety of explanations and examples for complex concepts

Customer gained a competitive edge by offering round-the-clock, personalized support to students across all subjects, which had a significant positive impact on student engagement.
Students of all age groups benefited from instant access to tailored explanations and assistance, allowing them to learn at their own pace and receive help whenever they needed it, regardless of time or topic complexity.
Teachers could focus on higher-level instruction and personalized guidance, as the AI chatbot handled routine questions and provided in-depth, holistic explanations.

AI-powered tooltips save hours of searching for materials and free up valuable time for more impactful educational activities and deeper student interactions.

The introduction of the tooltips feature makes the entire system independent of third-party vendors, as all information is retrieved from an internal database.

In addition, the solution was designed with flexibility in mind, allowing for seamless transition between cloud and on-premise deployments. This architecture ensures that, if required, migrating from cloud to on-premise infrastructure can be accomplished with minimal effort and disruption.

Overall, these initial positive results have opened up exciting prospects for further collaboration with the customer.

24/7	availability for student support
>1000	requests per minute handles the system
5+	points the customer grew the platform’s NPS

Aristek provided us with much-needed technology-specific expertise and support in the AI-based system implementation.

We were surprised at how fast it was possible to get such a project up and running. Also, it turned out to be not as difficult as it seemed at first glance, and we can already track positive changes in the engagement levels of students of all ages.

This solution became an all-rounder for inquisitive whyers worldwide, and we look forward to continuing to work on this project with the Aristek team.”

The customer’s product manager says

Want to know the cost of your AI project?

Answer a few quick questions to get an estimate.

Costs calculator

Contact our expert

If you have any questions regarding the case study or have similar project to be discussed, do not hesitate to contact us.