How Eklavya Is Rethinking AI Chats Without Breaking What Already Works

Building AI for serious exam prep is very different from building a casual chatbot.

At Eklavya.io, our goal is to help students prepare for JEE and NEET, exams where concepts build over months and accuracy really matters. Students do not just ask random questions. They study, revise, forget, return, and improve slowly over time.

Early on, we used RAGFlow to power our chats. It worked well. But as more students started using Eklavya, we realized something important.

The problem was not the answers.
The problem was control.

This blog explains what changed, why it mattered, and how we are fixing it in a way that keeps things powerful but clean.

---

How Things Worked Earlier

In the beginning, Eklavya used RAGFlow almost end to end.

Behind the scenes, we had NCERT books and exam-relevant material stored in RAGFlow. When a student asked a question, RAGFlow would:

• Pull the right NCERT context
• Keep track of the chat
• Generate the answer

From a user point of view, it felt simple. Ask a question, get an answer.

And honestly, this setup helped us move fast in the early days.

---

Where It Started Feeling Messy

As more students used Eklavya for longer periods, cracks started showing.

Chat memory is not learning memory

A normal chatbot remembers what you said five messages ago.
A learning platform needs to remember very different things.

For example:

• Which topics a student struggles with
• What level of explanation works best
• What concepts they keep mixing up
• How their understanding improves over time

Chat history alone is not enough for this.

---

Too much stuff going into the model

RAGFlow was doing two jobs at once:

• Bringing in NCERT context
• Holding chat memory

Over time, everything started mixing together. Context, chat history, small talk, follow-ups. The AI knew a lot, but not always the right things.

This made answers feel cluttered and harder to control.

---

Limited customization

For JEE and NEET, the same question can need different styles of answers:

• Short exam-oriented explanations
• Step by step derivations
• Intuitive understanding
• Quick revision mode

When chat logic lives inside a third-party system, adding these flavors becomes difficult.

We wanted freedom to shape the learning experience.

---

The Key Realization

We never stopped trusting RAGFlow.

RAGFlow was doing its job perfectly by giving us clean, syllabus-aligned context from NCERT books.

The real insight was this:

RAGFlow should handle knowledge retrieval.
Eklavya should handle conversation, behavior, and memory.

So instead of removing RAGFlow, we changed how we use it.

---

The New Way Things Work

Here is the new flow in simple terms:

1. A student asks a question on Eklavya
2. Eklavya decides what kind of help is needed
3. Eklavya asks RAGFlow only for relevant NCERT context
4. RAGFlow sends back focused, clean content
5. Eklavya builds the final prompt
6. The AI generates the answer
7. Eklavya updates what it knows about the student

Same high quality context.
Much better control.

---

What RAGFlow Still Does

RAGFlow remains a core part of Eklavya.

It is responsible for:

• Storing NCERT and exam material
• Finding the most relevant content
• Supplying trusted academic context

It no longer:

• Manages chat sessions
• Decides what to remember about students
• Controls answer style or behavior

Think of RAGFlow as a very smart library, not the teacher.

---

What Eklavya’s Own Chat System Does

By building our own chat and memory layer, Eklavya can now:

Remember what actually matters

Instead of remembering every message, Eklavya remembers:

• Topics studied
• Weak areas
• Preferred explanation styles
• Learning progress over time

This is real learning memory, not chat memory.

---

Change how answers feel

The same concept can now be explained differently based on:

• JEE or NEET focus
• Beginner or advanced level
• Revision or learning mode

This makes the experience feel more like a personal tutor.

---

Keep things clean

Only useful academic context goes to the AI.
Only meaningful learning data is stored about the student.

No clutter.

---

Why This Matters for JEE and NEET Students

Competitive exam preparation is a long journey. Students do not need a chatbot that remembers jokes or small talk. They need a system that remembers how they learn.

By separating:

• Knowledge retrieval
• Conversation flow
• Student memory

Eklavya becomes more reliable, more personal, and easier to improve over time.

---

Final Thoughts

Eklavya is not moving away from RAG.

We are simply using it more intentionally.

By keeping RAGFlow for context and building our own chat and memory system, we get the best of both worlds:

• Trusted NCERT-based answers
• Full control over learning experience
• Long-term student memory
• Freedom to evolve

At the end of the day, our goal is simple.

Help students understand better, remember longer, and feel more confident walking into their exams.