Eduboost

• 

  Edu Boost

Inspiration

Our family, in collaboration with an NGO supporting children from single-parent households affected by unforeseen events (e.g., farmer suicides, disease, parental incarceration), aims to provide education for both children and their parents through vocational courses. We observed that existing online courses are ineffective due to high dropout rates. Our research identified key reasons for this:

Declining Motivation: Initial enthusiasm quickly wanes.

Lack of Interaction: Students feel isolated without support.

Slow Support: Difficulty getting quick answers to queries.

No Rewards: The journey is entirely self-motivation dependent.

Therefore, the core reasons for higher dropout in online courses are the need for Motivation through interaction, Quick Support, and Rewards.

What it does

So, core reason of higher dropout in online courses are (A) Motivation and (B) Quick (C) Lack of quick support. And Eduboost helps to resolve this, by having education portal with hybrid approach of formal course content + Agent.

Three types of AI Agents will orchestra this learning journey by the students.

(a) Course Agent - Once course is enrolled, this guy is available at click far. This helps to respond to quick query and as it has all the information about the course content and course information. This agent has RAG index data of the course. It also has access to create support ticket.

(b) Content Agent - This has been designed for specific task like introduction, some short term interaction about the encouragement and interview to this is part of course content or course workflow.

(c) Quiz Assessment - Quiz creation is automated by the LLM. Student can take the quiz and on submit quiz is assessed by LLM and it provides structured feedback. When video interaction enabled for the quiz result agent counsel in a structured way - Strengths, Improvement and Suggestion

So, in a summary it provides end to end course generation, course enrollment, support or query ticket generation, taking quiz, getting result and finally seeing counselling on the result that got generated in a real time.

How we built it

Coding has been done 77 % with bolt.new, 1% - with other for fixing error, 1% - manual change for build errors. 1% - merging different modules, 17% - efforts in finetuning agents and researching on interactions with agents. 3%- Supabase database setup and resolving supabase edge functions.

We have been very concious about how to use the token effectively and effeciently, so thought twice or thrice before shooting for the prompt. We have selected modular approach. >Scaffolding So, we have developed first SAAS boilerplate, and make it stabilized. even that was step by step - created first scafolding, then added supabase database for authentication and social registration. I have created that as module so that if there is change required in basic structure like adding some functionality Module link : Github - Harshaldoshi/ReactSAASBoilerplate

>Course Management Then added core functionality of Adding the course, and enrollment, and uploading course content. It started with browser based storage, but then supabase addition made this seamless.

>Chapter or Milestone Agent - Motivates with video conversation: Tavus.io Agent Now its time to add agents. So, first added tavus.io. This agent can be added as a content of the chapter and can be assigned specific tasks. Replica id and Context are required to be added as part of the content data. >Course goto Agent - Quick goto Agent for any query: Elevenlabs.io Agent Another, thing that we wanted is Elevenlabs.io agent for entire course. 'Elevenlabs' can be mapped, to the course during course addition. This agent has entire course details as RAG, so any query from the user can be resolved by (LLM + Elevenlabs + Course content RAG). Module link : Github - Harshaldoshi/AgentQuickStart

>Ticket generation and 3rd party tool - Hotspot Integration : Elevenlabs.io Agent tools and agent transfer Wanted to enable support that user can flag a situation that he need some support and he could not resolve by course content or Agent. This can be done by agent itself by just saying "Wanted to generate ticket"

Quiz Generation on prompt, Evaluation and Counselling : _LLM Integration(tested with Gemini) and Tavus.io for counselling. Quiz generation is using LLM and automated. Option to add reference document from which quiz can be created or use LLM source. Counfigurations - Total marks, type of questions, LLM - Gemini, GROQ, OpenAI.

Challenges we ran into

Overall journey with bolt.new was good. There were few hurdles on solving the error. Resolving build error was a terrible experience, which I could solve in 5 min, but took lot of tokens by bolt without resolving it. Some technical glitch, while integrating Elevenlabs.io agent, it could not figure out conversational id. I could save lot of token, as I was working with modular approach, but I need to manually tweak some changes and then it took over nicely. Somehow, I could only work with widget provided by Elevenlabs.io. Even gemini could not resolve it, so some action item on Elevenlabs. While integrating, Gemini LLM, even though API were configured it correctly with edge functions, it was asking for the API on frontend. but then resolved the conflict.
UX settings for CourseEnrollment page, as there are multiple product.

Accomplishments that we're proud of

Structured and modular approach that we took helped to deliver good functionality with less token. (I willl be creating one article on medium on step by step journey), as I have all the documentation available with me. Elevenlabs agent with hotspot integration, agent transfer and RAG support.
Unlock of rewards feature - 'Special Download'

What we learned

I am first time user of React and Supabase, I guess this is oneliner to say that with bolt.new, one can develop a working full fledge system and not only rapid prototype without expertise. Also, I learnt of firing hubspot by invoking tool from Elevenlabs. This has gain my interest on MCP and more agent-agent conversation.

What's next for Eduboost

1. Automatic course creation - Prompt. (Already done with modular approach, will integrate it soon. )
2. Add subscription concept, where user can take multiple courses.
3. Adding apis for agent creation - Elevenlabs.io agents and Tavus.io from the admin panel itself. Metering of the Agent usage on tenant level / end-user level.
   
4. Create documentation for the Opensource code.
5. Create cloud paid version for Monetization. Free forever for NGO.
   
6. Add monetization from Quiz.

Built With

• daily.js
• elevenlabs-convai
• gemini
• groq
• llm
• lucide-react
• openai
• postcss
• react
• react-hot-toast
• react-router-dom
• supabase
• supabase-edge-functions-(deno)
• tailwind-css
• tavus.io
• typescript
• vite
• web-speech-api