Most AI chatbots answer from everything they have seen on the internet. That is fine for general questions — it is risky for board exams, where wording, diagrams, and definitions need to match your textbook.
Lerno does something simpler to describe: it searches your NCERT like a smart index, pulls the best-matching passages, and only then asks the AI to explain using that material. That pattern is called RAG (retrieval-augmented generation): retrieve first, generate second.
Built for NCERT, not the whole web
Your session is tied to grade, subject, and usually a chapter. So the tutor is not mixing random sources or another board’s syllabus — it stays inside the book you are actually studying.
The goal is not a wall of pasted textbook text. It is an answer that lines up with NCERT and still reads like a tutor talking to you.
From your message to the answer
Roughly this happens in order:
-
Scope — Your question is handled in the subject/chapter (or learn session) you picked, so search stays in the right part of the book.
-
Task shape — Lerno notices whether you want something like a normal explanation, short notes, a quiz, a problem worked out, or a summary — so the reply format matches what you asked.
-
Search query — Your message is cleaned up into a clearer search query (typos, casual phrasing, shorthand).
-
Embeddings — That query is turned into a numeric “meaning fingerprint” so the system can find relevant paragraphs even if your words do not match the book word-for-word.
-
Hybrid search — Lerno searches the textbook index in two ways at once: by meaning and by keywords (names, dates, exact NCERT terms). Results are merged so neither signal is ignored.
-
Ranking — Overlaps are removed and the best chunks are chosen so the model gets a tight bundle of context, not a pile of repeats.
-
Length — A quick guess decides if you need a short answer, a normal one, or a deeper one — so a small doubt does not turn into an essay.
-
Instructions + you — A system prompt keeps answers faithful to what was retrieved. Subject rules nudge format (for example maths vs history). Your memory (weak topics, pace, preferences from onboarding and usage) personalises the tone and emphasis.
-
Generation — A capable model (today, Google Gemini in Lerno’s setup) writes the answer, usually streaming so you can read as it appears. Photos or PDF pages can go to a vision model when you attach them.
-
Citations — When it makes sense, you see numbered markers like [1] or [2] that point back to chapter/topic/page-style source info so you can check the book.
-
Save — The reply and citations are stored with your chat so you can revisit them later.
In one line: find the right pages → pick answer style → search the book → rank → add rules and your profile → write → cite → save.
Ask, Learn, and Study Feed
Ask Mode — You ask whatever you want, one message at a time; the steps above run each time.
Learn Mode — Same engine, but the app walks you through a chapter in order so you are less likely to skip gaps.
Study Feed — Quick practice cards. How you do there also feeds the same mastery picture as chat, so “weak topics” are not only what you said you found hard — they can reflect what you actually missed in practice.
What Lerno is not
Lerno is a study helper, not a way to skip thinking or submit work that should be your own. The AI can still slip up (wrong retrieval, ambiguous question, model mistake), so always cross-check definitions and laws against NCERT, especially when marks depend on exact phrasing.
Your teacher and your textbook stay the final authority. Lerno is there to get you to “I get it” faster — with the book still in the loop.