AWS Only

Anthropic Models UsedEducationGen AI

Personalising Language Learning at Lingoda with a Generative AI-Powered Chatbot

Lingoda, an online language learning platform, partnered with Firemind to build a generative-AI powered chatbot using Amazon Bedrock's Large Language Models (LLMs). This solution simulated real-life conversations, provided constructive feedback tailored to proficiency levels, and enhanced learners' skills through immersive role-play scenarios and personalised assessments. The goal was to deliver interactive and engaging English language courses to Lingoda's users.

AWS ONLY ·

Personalising Language Learning at Lingoda with a Generative AI-Powered Chatbot

Lingoda, an online language learning platform, partnered with Firemind to build a generative-AI powered chatbot using Amazon Bedrock's Large Language Models (LLMs). This solution simulated real-life conversations, provided constructive feedback tailored to proficiency levels, and enhanced learners' skills through immersive role-play scenarios and personalised assessments. The goal was to deliver interactive and engaging English language courses to Lingoda's users.

At a glance


Lingoda is an online language learning platform that offers online language courses with live, interactive classes taught by native speakers. Their platform provides flexible schedules, personalised learning paths, and a range of languages, enabling learners to improve their skills and achieve fluency.

Challenge

The need for interactive role-play with real-time, personalised feedback across different proficiency levels.

Solution

Developed an AI chatbot using generative AI for dynamic role-play and tailored feedback.

Services Used

Amazon Bedrock
Amazon DynamoDB
Amazon S3
AWS Lambda

Outcomes

Average reduced cost was reduced to $0.020 per session.
Increase in learner engagement with the AI-Powered chatbot.

Business challenges

Overcoming the limitations of traditional language learning

Lingoda aimed to advance its English language courses by incorporating interactive role-play scenarios that could provide personalised, real-time feedback. The challenge was to create a system capable of engaging students in dynamic conversations, tailored to various proficiency levels, while evaluating their performance accurately. Traditional methods lacked the necessary interactivity and adaptability, making it difficult to offer individualised learning experiences at scale.

The solution needed to integrate seamlessly with Lingoda’s existing platform, handle diverse scenarios, and support configurable learning outcomes. Additionally, it had to be cost-effective while maintaining high-quality interactions. Ensuring that the AI-driven system met the CEFR level requirements and provided constructive feedback throughout each session was critical to achieving Lingoda’s educational goals and improving learner engagement.

"Firemind's PULSE tool was able to quickly and accurately summarise vast datasets of customer surveys, ensuring we could take action against any negative customer experience elements."

Dean Macfadyen, Data Platform Engineer — Generative AI Customer

Solution

Building a conversational AI chatbot powered by Amazon Bedrock

Firemind developed a conversational AI chatbot powered by Amazon Bedrock’s Large Language Models that seamlessly integrated with Lingoda’s existing online language learning platform to address Lingoda’s challenge. The solution consisted of three agents – a greeter agent, a chat agent, and an evaluator agent – that worked together to guide the learner through role-play scenarios, provide feedback, and assess their performance based on the defined learning outcomes and CEFR levels.

The chatbot was designed to effectively engage learners in interactive conversations, tailored to their proficiency level. The greeter agent would introduce the scenario and the bot’s role, the chat agent would respond to the learner’s inputs and provide constructive feedback, and the evaluator agent would assess the learner’s progress across key criteria like task achievement, vocabulary usage, grammar, and appropriateness.

This AI-driven solution allowed Lingoda to move beyond traditional language learning methods and offer their students a more immersive, personalised practice environment. By leveraging the capabilities of Large Language Models, Firemind was able to create a scalable, adaptive system that could simulate realistic conversations and deliver detailed, real-time assessments to support Lingoda’s language education goals.

Improved learner engagement

The conversational AI chatbot successfully engaged learners in role-play scenarios, fostering a more immersive and interactive language learning experience. By simulating realistic conversations and adapting to the learner's proficiency level, the solution helped keep students motivated and actively practicing their skills.

Optimised costs

Firemind's use of AWS services, including Amazon Bedrock, DynamoDB, and Lambda, enabled a cost-optimised implementation of the conversational AI chatbot. By leveraging the Sonnet 3.5 model, the team achieved a reduced cost of $0.020 per session, optimising Lingoda's budget and resource allocation.

Model Spotlight


Claude 3.5 sonnet

We chose Claude Sonnet 3.5 for the Lingoda roleplay chatbot because of its ability to adapt based on a student’s language proficiency level, offering real-time feedback and matching the intensity of the module to the learner’s needs. Its strong roleplay capabilities allowed it to effectively take on the role of a tutor, guiding students through lessons with a conversational, gamified approach.

In earlier tests with Claude 3, we encountered challenges where the model struggled to follow instructions, requiring extensive prompt engineering to stay on task. However, Claude 3.5 showed a marked improvement, significantly reducing the need for prompt engineering while providing more natural and contextually accurate feedback.

This led to cost savings and improved the overall learning experience by keeping conversations aligned with the student’s level and goals. While other models like Claude Haiku were considered, they fell short in comparison, making Claude 3.5 the optimal choice for this project.

Learn more

Why Firemind

Lingoda selected Firemind for its expertise in AI-driven solutions and innovative approach to language learning. Firemind’s proven track record in developing advanced, generative AI technologies and seamless integration capabilities offered Lingoda a cutting-edge solution for enhancing their platform with interactive role-play and real-time feedback, addressing their specific needs effectively.

1920%

Workflow speed increase

1,000 documents in 12.5 minutes compared to 4 hours by human operator

88%

Keyword accuracy

for leading categories within data modelling & training

Added value

We were also able to take all stakeholders on a journey concerning how Cloud Adoption and Machine Learning would benefit their customers and their business continuity, all in a 12 week timeline from start to finish.

Get in touch

Want to learn more?

Seen a specific case study or insight and want to learn more? Or thinking about your next project? Drop us a message below!