TheGuides

Workflow
TiDB Metric
Home and Response Sample
Response with new Subject endpoint
"query: explain hypothesis testing" - Translated in Hindi
Chat History Tab
Responds within the scope of subject

Inspiration

Inspired by GPT, the power of AI and LLM, evolving to an interactive learning experience.
Motivated to increase the quality of educational access across all regions and languages.
From students perspective, how we asking the right question is really important for us to learn directly from the relevant and useful resources in a language and difficulty as per our requirements.

We want to maximize the performance and scalability of TheGuides using TIDB allowing everyone in the world, including all ages and backgrounds, to have access to this interactive learning experience. We believe this is the innovation to guide people how to learn better with AI and our mission is to deliver this interactive learning method with agility and scale.

What it does

TheGuides is an AI-powered educational chatbot that helps users to focus and learn better from a role-played expertise (any sector) that is trained with a data layer.
Supports over 100+ languages to educate users better than ever.
Sets up a personalized educational mentor.

How we built it

We are building on top of OpenAI’s LLM with langchain, using the most recent GPT model to experience how we should learn in a brand new way.

Using the vector embedding property of GPT to save custom data from user input to enhance the user experience.
Storing user’s questions and AI’s responses in TiDB’s database.

Challenges we ran into

Server, database, and LLM were completely new to us, some example shown below:

SOLVED

Building the server with FastAPI and integrating the vector database along with MySQL database on TiDB.
New function calling capability of GPT.
Connect the server and frontend through API calls.
Built locally using SQLalchemy and SQLlite then switching to MySQL to integrate with TiDB.
Connecting to TiDB while working on different OS caused certification errors.

PROBLEM

Deployment of the project on Vercel caused HTTP 508 error.
Adding individual subject choices are still a work in progress, works perfectly on server side but just a placeholder on the website for now.

We still managed to connect everything in one piece that potentially can help students and professionals.

Accomplishments that we're proud of

We successfully deploy an end-to-end working personalized guidance chatbot using subject as role play and career expertise data.

What we learned

We understood more about LLM and how the whole backend server and frontend work as a whole, as well as, the prompting message for the chatbot as our key feature component.
We have also learnt the plus sides of using cloud based serverless databases over locally hosted databases. It provides improved security as well as easy access and uninterrupted availability.

What's next for TheGuides

We believe learning shouldn't be something people should set a time range nor stop after school. Our vision and mission are delivering interactive learning experience with AI & LLM would be the trend for the next decade and we want to guide people learning better, understand how to proceed questioning in a better sense, or do prompting better than ever in a transformed way.

Keys moving forward:

Users should be able to take full benefit of vector embeddings and be able to upload their study notes or university notes and ask personalized questions on it.
They would be able to select topics of their choice and then ask questions on them, could be a particular subject (such as English, Physics, Chemistry, History, Mathematics, etc.) or career-wise (such as Engineer, Financial Advisor, Lawyer, etc.)
how adults (over 25 years old) can also learn better.

As far, our working model is manually set up, however, we are looking forward to adding more datasets from different professionals from all sectors for a content creator community for educational evolution.

Update for Final Round

Modified the UI and format of chatbot response improving the user point of view
Be more precise on prompt engineering to set up the role of AI model for preparing more examples for the users
In progress of the subject button feature that allow users to trigger specific subject or role for the AI model
Successfully fetch subject value from front end to server and passing it to the AI model
Bot focuses on one subject at a time to keep the conversation cleaner and organized.

Suggestions

TiDB works flawlessly with the storage server but since it integrates perfectly with AI and LLM based applications, it should introduce vector embeddings in itself.
Adding support to other languages such as Golang and JavaScript will also be a great addition to the TiDB project