In March 2026, we filed a patent for an AI system that generates live visual lessons in real-time. We're two cofounders from VIT Vellore and SRM, running a 5-person team, with zero VC funding.
Here's how we built it.
The Problem We Were Solving
EaseLearn AI started as a doubt solver — point your camera at a question, get an instant solution. 50,000 students started using it.
But we kept hearing the same feedback: "Solving doubts is great. Can AI actually teach me?"
Teaching is fundamentally different from answering. Answering is reactive — you ask, it responds. Teaching is proactive — it structures information, builds on previous concepts, checks understanding, and adapts.
No AI product was doing this well. ChatGPT gives you text. YouTube gives you pre-recorded videos. Neither adapts. Neither teaches.
So we built Immersive Classroom.
The Architecture
Immersive Classroom has three core components:
1. Lesson Planner
When a student picks a topic, the planner:
- Breaks the topic into 5-8 logical segments
- Orders them by pedagogical best practice (based on Google's education research papers)
- Identifies which segments need diagrams, which need examples, which need practice questions
- Adapts the depth based on the student's grade level and past performance
2. Slide Generator (IRIS)
For each segment, IRIS generates:
- A presentation slide with a clear heading and structured content
- Live images — diagrams, charts, illustrations created in real-time (not pulled from a library)
- Embedded interactive elements (fill-in-the-blank, MCQ checkpoints)
The key technical challenge: generating visually coherent slides fast enough to feel "live." We optimized for sub-2-second generation per slide.
3. Voice Synthesizer (Veda AI)
Veda AI generates the verbal explanation synchronized with the slides:
- Natural language explanation of each concept
- Pacing matched to slide transitions
- Multilingual support — same lesson in English, Hindi, or any language
- Adaptive tone — simpler language for younger students, technical language for JEE/NEET prep
The Technical Decisions That Mattered
Why not just use GPT-4 and call it a day?
We tried. Raw LLM output is not teaching. It's answering. The difference:
- Teaching has structure — introduction, explanation, example, practice, summary
- Teaching has pacing — you don't dump everything at once
- Teaching has feedback loops — you check if the student understood before moving on
- Teaching has visuals — a diagram of the Krebs cycle teaches more than 500 words about it
We use LLMs as a component, not as the product. The orchestration layer — lesson planning, slide generation, adaptive difficulty, quiz embedding — is our IP.
Why real-time generation instead of pre-generated content?
Pre-generated content can't personalize. If you pre-generate a lesson on "Electromagnetic Induction," it's the same for a Class 12 CBSE student and a JEE Advanced aspirant. Real-time generation means:
- The CBSE student gets a simpler explanation with NCERT-aligned examples
- The JEE student gets a deeper explanation with previous year problems
- A student who got the first quiz wrong gets a re-explanation before moving on
- A student who got everything right skips ahead
Why patent it?
Because no one else is doing this. We searched. Plenty of AI chatbots. Plenty of video platforms. Zero products that generate live visual lessons in real-time with adaptive difficulty. The combination of real-time slide generation + live image creation + adaptive teaching + embedded assessment is novel.
The Stack
- Frontend: Next.js (web), React Native (mobile)
- Backend: Node.js on Google Cloud Run
- AI: Multi-model orchestration (we use Claude for reasoning, Gemini for multimodal generation)
- Infrastructure: GCP + AWS (Lambda for OCR pipeline, S3 for media)
- Real-time: WebSocket connections for live lesson streaming
What We Learned
1. Students don't want more content. They want better explanations.
India has infinite free content. What's missing is personalized teaching. The AI that wins isn't the one with the most data — it's the one that explains things the way you need to hear them.
2. Camera-first beats chat-first for Indian students.
Indian students study from physical books and handwritten notes. A chat interface that requires typing math equations is useless. Camera-based input was our single best product decision.
3. Free is the only viable price for student acquisition in India.
We tried freemium. Students don't convert. We made the core product free and added premium features (Immersive Classroom, unlimited doubts, PYQ solver) at Rs 199/month. Conversion rate tripled.
4. You don't need VC money to build AI products.
AWS Activate, Google for Startups, NVIDIA Inception — startup programs gave us the compute credits we needed. Our total infrastructure cost is under $2,000/month serving 50,000 students.
Try It
Immersive Classroom launches April 17. First 500 students get 1 month free.
If you're a developer interested in AI + education, we're hiring. If you're a student, try it. If you're a founder building in edtech, let's talk.
EaseLearn AI is a free AI study companion for Indian students. Doubt solver · AI tutor · Math solver · Immersive Classroom
Top comments (0)