The AI tutor that feels like your favourite teacher.
India has 248 million students and nowhere near enough attention to go around.
Eila is an AI teacher that notices when a student is lost, remembers what they struggled with last week,
and explains it again — patiently, in the language they think in.
Not a chatbot — a teacherEmpathy engineSyllabus-groundedHindi · Hinglish · English
A CWI Studio product
Bringing AI to the problems education actually has
E
Eila · Maths
listening
HINGLISH
sir yeh integration wala sum samajh nahi aa raha 😔
remembers · asks · adapts
Koi baat nahi — pichhli baar bhi substitution pe atke the. Ek chhota sa sawaal:
is expression mein aisa kaun sa part hai jiska derivative bhi expression mein dikh raha hai?
x² ka derivative 2x… aur 2x toh yahin hai!
Bilkul 👏 wahi hamara u hai. Ab tum khud likh ke dikhao — main dekh rahi hoon.
Student twin · Class 11
Substitution38%
By parts21%
Definite ∫72%
Ask Eila anything…🎙
Eila AI · CWI Studio01 · Cover
02 The Problem
India's education crisis is a scale problem.
There is no shortage of syllabus, textbooks or video lectures. There is a shortage of one thing that
learning cannot happen without: a teacher who is paying attention to this student, right now.
One teacher. Thirty students. One period.
🧑🏫
80 seconds
of individual attention, per student, per period
NEP 2020 sets 30:1 as the pupil–teacher benchmark. A 40-minute period divided thirty
ways is 80 seconds per student — and states such as Bihar still sit above that ratio (UDISE+ 2024-25).
Personalisation here is not a teaching-quality problem; it is an arithmetic one.1,2
24.8 Cr
Students
Enrolled in school, 2023-24 — the largest student population on earth1
14.7 L
Schools
Across India, foundational to secondary1
27%
Already pay
Of students take private coaching — 30.7% in urban India3
⏱️
Taught at syllabus pace, not student pace
The class moves when the calendar says so. A student who did not understand Chapter 4 is carried into
Chapter 5 anyway — and the gap compounds silently for years.
🙊
The doubt is never voiced
Asking in front of thirty classmates is socially expensive. The doubt does not disappear when it goes
unasked — the student does.
📼
EdTech gave content, not teaching
A recorded lecture cannot notice a confused face, cannot slow down, cannot ask "so what did you get?"
More content was never the missing ingredient.
Millions fall behind — not for lack of ability, but for lack of attention.
Eila AI · CWI Studio02 · Problem
03 The Human Truth
What the student actually feels is not a content gap.
We started from the student's side of the desk. Every story we heard came back to the same three feelings —
and none of them are solved by another video library.
😶
Fear
“I have doubts but I feel scared to ask.”
An AI teacher has infinite patience and zero social cost. In one classroom study,
93% of students who avoided asking questions said they would ask if it could be done anonymously.8
📺
Disconnection
“Videos don't explain like a real teacher.”
Teaching is a conversation with feedback. Eila asks back, checks understanding,
and changes its explanation when the first one doesn't land.
👤
Invisibility
“My teacher is too busy — I feel invisible.”
Eila remembers every student individually: what they got wrong, what finally
clicked, and how they like to be explained to.
Our thesis
Students do not need more content. They need to be noticed.
Eila AI · CWI Studio03 · Human Truth
04 The Gap
Everything on the market answers. Nothing teaches.
Three product categories already compete for the Indian student's screen time. Each one solves a different
problem from the one the student actually has.
Category
What it is
What it actually does
Why it does not teach
Content apps
Recorded video platforms
Streams a lecture at every student identically
Zero personalisation — the video cannot tell whether you understood it
Doubt apps
Photo-to-answer tools
OCRs the question, returns the solved answer
Answers, never teaching. The student copies; the reasoning never develops
Generic AI
General-purpose chatbots
Answers anything, very confidently
Not syllabus-mapped, no memory of the student, no sense of when to slow down
Eila AI
An AI teacher
Diagnoses, questions, re-explains, remembers
Fills the gap: personalised · empathetic · syllabus-aligned · multilingual
🎯
Answering ≠ teaching
Giving the answer ends the learning event. A teacher withholds it, asks the one question that unlocks the step, and lets the student arrive.
🧠
Stateless ≠ personal
A model with no memory meets a stranger every session. Learning is cumulative; the tutor must be too.
🌏
English-first ≠ Indian
Most students reason in Hindi or Hinglish and are forced to learn in English. Eila teaches in the language the thought actually happens in.
Eila AI · CWI Studio04 · The Gap
05 The Product
Eila AI — a complete teaching system.
Not a chatbot with a syllabus prompt bolted on. Four capabilities, each engineered as its own system,
that together add up to something a student experiences as a teacher.
🎯
Personalised
Adapts to every student's learning curve — pace, difficulty and revision cadence set per learner, never per class.
💛
Empathetic
Reads frustration and disengagement in the exchange, then shifts tone, slows down and encourages before continuing.
💬
Interactive
A conversation, not a lecture. Socratic questioning builds the reasoning instead of handing over the answer.
🌏
Multilingual
English, Hindi and Hinglish at launch; regional languages next. Students learn fastest in the language they think in.
Subject tutors with distinct personalities
📐
Maths
Calm, patient, step-by-step.
🧪
Science
Curious, energetic, real-world.
📖
English
Warm, story-driven, expressive.
Personality is not decoration. A student who finds their tutor relatable comes back tomorrow — and retention is the whole game.
Built for how students actually study
🎙Voice-first — speak the doubt, hear the explanation. Typing is a barrier for exactly the students who need help most.
🧑Avatar delivery — a face and a voice that behave like a person teaching. Presence is what separates a lesson from a lookup.
📸Ask by photo — snap a sum from the textbook and Eila teaches it, instead of just returning the answer.
📶Real phones, real networks — mid-range Android, low bandwidth, patchy connectivity. The devices students own.
Eila AI · CWI Studio05 · Product
06 Beyond Subjects
The greatest education ever devised was one tutor, one child.
Aristotle taught Alexander. Chanakya taught Chandragupta Maurya. For most of history the finest education on
earth was a single mentor shaping a single mind — and it was reserved for royalty.
Eila makes that the default for every child.
🧭
Thinking frameworks
First principles, cause and effect, breaking a hard problem into small ones — the skill that outlives the syllabus.
🦉
Wisdom & judgement
What to do when there is no formula — weighing trade-offs, knowing what you don't know.
🌱
Character & discipline
Consistency over intensity, honesty about mistakes, effort valued above talent.
⚖️
Morals & ethics
Right and wrong reasoned through with the child — the Socratic method applied to life.
Taught through the child's own world
Eila knows the student's town, their school, the shop on their street. Every lesson is grounded in people
and places the child actually knows — so it is imbibed, not memorised.
“Socho — tumhare mohalle ki kirana dukaan. Sharma ji udhaar kyun dete hain, jab pakka nahi hai ki paisa
wapas aayega?”
Why one-to-one wins — the 2σ effect
Conventional classroomOne-to-one tutoring
Bloom, 1984 — the average tutored student outperforms 98%
of classroom students. A 40-year-old finding nobody could afford to deliver. Inference costs just made it a subscription.6
Then · 343 BCE
Aristotle → Alexander
Then · 321 BCE
Chanakya → Chandragupta
Now
Eila → every child
Eila AI · CWI Studio06 · Beyond Subjects
07 Inside a Session
Four things happen that no chatbot does.
A single exchange, unpacked. This is the difference between a tool that answers and a teacher that teaches.
1It notices the feeling first. “😔” is data. Eila registers frustration and opens with reassurance before it opens with content.
2It recalls this student. “Pichhli baar bhi substitution pe atke the” — the Student Twin remembers the exact concept that broke last time.
3It asks instead of answering. The Socratic step hands the reasoning back to the student, so the next sum can be solved without Eila.
4It speaks their language. Hinglish is not a translation feature — it is how the student's own thinking actually sounds.
Mastery updated liveConcept traced to the syllabusEscalates to a human when stuck
E
Eila · Maths
teaching
HINGLISH
sir yeh integration wala sum samajh nahi aa raha 😔
empathy → recall → question
Koi baat nahi — pichhli baar bhi substitution pe atke the. Ek chhota sa sawaal:
is expression mein aisa kaun sa part hai jiska derivative bhi expression mein dikh raha hai?
x² ka derivative 2x… aur 2x toh yahin hai!
Bilkul 👏 wahi hamara u hai. Ab tum khud likh ke dikhao — main dekh rahi hoon.
Student twin · updated
Substitution54%↑
By parts21%
Definite ∫72%
Ask Eila anything…🎙
Eila AI · CWI Studio07 · Experience
08 How It Works
An intelligent teaching pipeline — not a prompt.
Every question is routed through a teaching policy, grounded in the board curriculum, personalised against
that student's own profile, and then delivered as voice and avatar. The model is one component, not the product.
01
💬
Query
Student asks a doubt — typed, spoken or photographed.
02
⚙️
Tutor Policy
Decides how to teach: hint, Socratic question, worked example or re-explain.
03
🕸️
Curriculum Graph
Grounds it in the board syllabus — chapter, concept, prerequisite chain.
04
🪞
Student Twin
Applies this learner's history: past errors, strengths, pace, language.
05
🗣️
Response
Retrieval-grounded generation → voice + avatar → an answer that teaches.
Retrieval-grounded · syllabus-citedModel-agnostic coreLow-latency streamed responseOptimised for mid-range AndroidPer-student data isolation
🔒
Grounded, not guessing
Retrieval over board-mapped curriculum content keeps every answer inside the syllabus and traceable to a chapter — the failure mode a general chatbot cannot fix.
🧬
Memory that compounds
The Student Twin persists across sessions and subjects, so each interaction makes the next explanation sharper for that specific child.
🧩
The teaching layer is ours
Frontier models will keep improving and getting cheaper. The policy engine, curriculum graph and twin sit above them — we benefit from every model release instead of competing with it.
Eila AI · CWI Studio08 · Architecture
09 What Makes It Different
Five things a general chatbot structurally cannot do.
These are not prompt tricks. Each one is a system we design, build and improve with every student interaction.
🧠
Context-aware memory
Remembers mistakes, strengths and pace across sessions — the way a real teacher remembers a class.
❓
Socratic method
Asks the question that unlocks the reasoning instead of handing over the final answer.
💛
Empathy engine
Reads sentiment in the exchange and adjusts tone, pace and encouragement accordingly.
📈
Adaptive difficulty
Scales problem complexity in real time so the student stays in the productive-struggle zone.
🌏
Multilingual teaching
English, Hindi and Hinglish — tuned for how Indian students actually speak and reason.
We measure learning, not screen time
Engagement metrics reward addiction. Eila is graded on
concept-mastery lift and doubt-resolution rate — if a student needs us
less on a topic next month, the product worked.
Teacher-aligned, not teacher-replacing
Eila takes the doubts a teacher cannot reach at thirty-plus to one and hands
back a live map of who is struggling and where. The classroom stays human; the attention gap closes.
Eila AI · CWI Studio09 · Differentiation
10 Why Now
This product was impossible eighteen months ago.
Four independent curves crossed at the same time. A one-to-one AI teacher for an Indian student is now
technically possible, culturally accepted, and — critically — affordable.
The unlock: inference cost collapse
$ per million input tokens, frontier-class models — log scale. Indicative published list
prices. A ~200× fall in three years is what turns a ₹299/month tutor from a
subsidised experiment into a positive-margin product.7
📱
1.03 billion online
India passed a billion internet subscribers in Dec 2025 — 983M of them wireless. The delivery device is already in the student's hand.4
📜
NEP 2020
National policy mandates personalised, competency-based learning. The tailwind is regulatory.
🗣️
Indic language models
Hindi and Hinglish crossed the threshold where a model can teach in them, not just translate.
🌐
#2 globally
India is the second-largest consumer of digital learning. Demand is proven; teaching is the missing supply.
⚡ The window is open, briefly
Global players optimise for English-first, general-purpose assistants. The India-specific teaching layer is
unbuilt — and whoever builds the curriculum graph and student memory first, keeps it.
Eila AI · CWI Studio10 · Why Now
11 Market
A generational market — and a narrow way in.
We are not trying to serve 248 million students on day one. We are starting where the pain, the willingness
to pay and the measurable outcome all peak — and expanding outward from there. Sizing below is IBEF / IAMAI
published market data; SAM and SOM are our own estimates, marked as such.5
Indian EdTech 2030 forecast
₹2,50,850 Cr≈ $29B
Indian EdTech today
₹64,875 Cr≈ $7.5B
SAM · K-12 exam-prep our segment
₹30,000 Crest.
SOM · 5-year target
₹900 Crtarget
~25%
CAGR implied, 2024→20305
Beachhead
Grades 9–12, preparing for Boards / JEE / NEET
The highest-stakes years, the highest willingness to pay, and the sharpest measurable outcome —
a mark on a scoreboard everybody already trusts.
🏙️
Then Tier 2 & 3
Where private tuition is least available and the attention gap is widest — Eila's advantage grows as you move away from the metros.
🗺️
Then regional
Five more Indian languages, then state boards. The same engine, retargeted per curriculum.
Eila AI · CWI Studio11 · Market
12 Defensibility
Models get commoditised. Context does not.
Anyone can call the same model we call. Nobody can call our curriculum graph or a student's twin.
Every asset below gets harder to replicate the longer Eila runs.
01
Proprietary curriculum graph
Board syllabi mapped concept-by-concept with prerequisite chains — structured domain work, not a scrape.
02
Student twin memory
A persistent learning profile per student. Switching cost rises every session — the tutor that knows you is hard to leave.
03
Teaching policy engine
The rules for how to teach a given student a given concept — refined on real outcomes, not on model releases.
04
Indian-language tuning
Hindi and Hinglish teaching quality that global, English-first models do not optimise for.
05
Data flywheel
Every interaction improves the policy engine, which improves outcomes, which wins the next student.
Flywheel
More students → more teaching interactions → a better policy engine → measurably better results →
more students, at lower cost of acquisition.
🛡️
Trust & child safety
India-hosted data, per-student isolation, age-appropriate guardrails and full transcript auditability — the posture parents and schools require before they let AI near a child.
🧱
Model-agnostic by design
The teaching layer is decoupled from any single provider. When a better or cheaper model ships, Eila gets better and cheaper with it.
Eila AI · CWI Studio12 · Moat
13 Business Model
Affordable for a family. High-margin at scale.
27% of Indian students already pay for private coaching — ₹3,988 a year in urban households, ₹9,950 at
higher-secondary. Eila's entry plan costs less, covers every subject, and is awake at midnight.3
B2C · Primary
Student subscription
₹299 – ₹1,999 / month
Free tier to acquire; premium unlocks full subjects and voice
Multiple children, all subjects, one household plan
Parent view: what each child is struggling with this week
Higher retention and materially better unit economics
B2B · Expansion
School licensing
₹50 – ₹200 / student / month
Bulk seats, sold to the school rather than the parent
Teacher dashboard: who is struggling, on what, right now
Low-touch distribution once the B2C outcome is proven
Annual spend per student — what a household already pays
Coaching · urban higher secondary
₹9,950
Coaching · urban all levels, average
₹3,988
Eila · entry plan ₹299/mo, all subjects
₹3,588
Eila · school licence ₹50/student/mo
₹600
₹0₹3,000₹6,000₹9,000
₹ per student per year. Coaching figures are NSS 2025 national averages.3 Eila's
entry plan lands below what an urban household already spends — and covers
every subject rather than one.
70%+ target gross margin, widening
Inference is the only meaningful variable cost — and it falls every quarter while price holds. Caching, routing and distillation cut it further as usage patterns become known.
Low CAC, high retention
Educator advocates and exam-prep communities first; paid acquisition only once retention is proven. An academic year is nine months long — a tutor that remembers you is a subscription you do not cancel.
Eila AI · CWI Studio13 · Model
14 Go To Market
Low-cost, high-trust — outcomes do the marketing.
Education is bought on trust, not on advertising. We prove the teaching quality on a small cohort,
let the results travel, and only then spend to scale.
Phase 1
Closed beta cohort
A few hundred students, one board, three subjects. Instrumented end-to-end so we can prove mastery lift, not usage minutes.
Phase 2
Educator advocates
Exam-prep creators and teachers who already hold student trust. Their endorsement converts at a fraction of paid-ad cost.
Phase 3
School partnerships
Tier-2 and Tier-3 schools take bulk seats once the outcome data exists. One signature reaches a thousand students.
Phase 4
Regional expansion
Additional Indian languages and state boards on the same engine — the graph is rebuilt, the product is not.
🧪
What we de-risk first
Teaching quality on a deliberately narrow slice — one board, three subjects, two languages — before any spend on scale.
📊
What we measure
Concept-mastery lift, doubt-resolution rate, weekly active learners, and retention into the next term.
👪
Who signs the cheque
The parent buys, the student uses, the teacher endorses. All three have to feel the product working.
🗣️
Why it spreads
A student whose marks move tells their friends. In exam-prep circles, that is the only distribution that has ever mattered.
Eila AI · CWI Studio14 · GTM
15 Execution
An 18-month path with no unknown physics.
Every component below is buildable with technology that exists today. The risk is execution and teaching
quality — not research.
Year 1
✓Q1–Q2 — curriculum graph, tutor policy engine, student twin. MVP in Hindi / Hinglish.
✓Q3 — voice tutor and avatar delivery. First 5,000 active students.
✓Q4 — JEE / NEET modules, parent view, first school pilots.
Year 2
✓Scale — target 100,000 active students.
✓Expand — five additional regional languages.
✓Deepen — full school deployments with teacher dashboards.
Where the first cheque goes
🧠
Core engineering
The teaching policy engine, curriculum graph and student twin — the three assets that are ours forever.
📚
Curriculum mapping
Subject experts encoding board syllabi concept-by-concept. Slow, unglamorous, and the hardest part to copy.
⚙️
Inference & infrastructure
Serving the beta cohort at low latency on real devices, and instrumenting every session for outcome measurement.
🎓
Pilot & measurement
Running the closed cohort, paying for the assessment that proves mastery lift is real.
Eila AI · CWI Studio15 · Roadmap
16 The Studio & The Vision
CWI Studio — building for how students learn.
CWI Studio
AI & Technology Studio
A product-first AI studio built on one conviction: education does not need more software that manages
students — it needs software that teaches them. We pick problems where AI
changes the outcome, not just the interface. Eila AI is our first.
5+ years shipping products across FinTech, EdTech and Web3 — backend systems,
DevOps and applied AI. IIT Kharagpur; Google Summer of Code alumnus ('21 & '22); published researcher.
Building Eila as a founder who writes the code, talks to the students, and owns the outcome.
Every student deserves a teacher who remembers their name, their pace, and their last mistake.
Kings bought that for their heirs. Everyone else got a classroom of forty. Eila makes it the default —
not a cheaper school, not a faster video, but a patient, attentive mentor for
every one of India's 248 million students: the subject, the thinking behind it,
and the character to use it well — in the language they think in, at a price a family can actually pay.
🎯
Teach, don't answer
💛
Notice, don't stream
🇮🇳
Built for India first
Eila AI
The AI tutor that feels like your favourite teacher
Let's build it together
CWI Studio Motihari, Bihar · India
Eila AI · CWI Studio16 · Thank You
17 Appendix
Sources & assumptions.
Every figure in this deck is either published data with the source named below, our own arithmetic from
those figures, or a stated target. We have marked which is which.
Published sources
1UDISE+ 2023-24 & 2024-25, Ministry of Education, Govt. of India — 24.8 crore students enrolled; 14.72 lakh schools; pupil–teacher ratio 10:1 foundational to 21:1 secondary, with Bihar among the highest.
3Comprehensive Modular Survey: Education 2025 (NSS 80th round), MoSPI — 27.0% of students take private coaching (30.7% urban); average annual coaching spend ₹3,988 urban, ₹9,950 urban higher-secondary.
4TRAI Telecom Performance Indicator Report, Oct–Dec 2025 — 1,028.61 million internet subscribers; 983.29 million wireless.
5IBEF / IAMAI — Indian EdTech ₹64,875 Cr (US$7.5B) today, forecast ₹2,50,850 Cr (US$29B) by 2030; ~25% CAGR implied.
6Bloom, B. S. (1984), “The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring”, Educational Researcher 13(6), 4–16.
8Montgomery, Evans, Harrison & Damian, “Towards a Live Anonymous Question Queue To Address Student Apprehension”, Univ. of Victoria — classroom study of question apprehension (undergraduate, non-India).
Our estimates — not published data
7Inference cost curve — 2023–24 points are published list prices for frontier-class models; 2025–26 are our extrapolation, shown as indicative.
·SAM and SOM — our segmentation of the IBEF/IAMAI total, not a published sub-market figure.
·Pricing, gross margin and user targets — our plan, not forecasts of demand.
Derived arithmetic
“80 seconds per student” is a 40-minute period divided by NEP 2020's
30:1 benchmark. “₹3,588 a year” is ₹299 × 12. Both are shown so a reader can check them.
