Education · Voice AI
Pathshala AI
A patient, voice-first tutor in every student's own language.
An AI tutor that explains the national curriculum in spoken Bangla, grades handwritten work from a phone photo, and adapts to each student's weak topics. Priced for a family budget, not a coaching-centre budget — and built to travel to any exam system in any language.
Launch market: Bangladesh · 4 cited sources · Facts reviewed 14 October 2026
- Cash to MVP (est.)Our estimate of the cash needed to reach a working MVP: one-time setup plus monthly running costs for the build period. It excludes founder salaries.
- $3,400
- MVP timelineMinimum viable product: the smallest version real users will use and pay for, built to learn fast.
- 6 weeks
- Margin after AI (est.)Of every 100 you charge, how much is left after paying for the AI models and infrastructure behind it. Higher means a healthier business.
- ~70%
- Editorial scorePlankito's own judgement: the average of six 1–5 ratings (AI leverage, moat, competition gap, capital efficiency, timing, scalability). It is an opinion to compare ideas, not a forecast.
- 4.3 / 5
At a glance
- Who uses it
- School students preparing for SSC and HSC exams, and their parents.
- Who pays
- Families, by monthly subscription; private schools later.
- Who should build it
- A product engineer plus a teacher who knows the national curriculum.
- The hardest part
- Answers must be right. One confident wrong answer costs a family's trust.
01
Opportunity
Families spend heavily on private coaching because classrooms are crowded and teachers cannot give one-to-one attention. Students in smaller towns get the worst of it: fewer good tutors, the same exams. The world is short of teachers, and the shortage is deepest exactly where families can least afford private help.
The evidence
70%
Globalof 10-year-olds in low- and middle-income countries cannot read and understand a simple text.
44M
Globalmore teachers are needed worldwide by 2030 — 15 million of them in sub-Saharan Africa alone.
265M+
Regionalpeople speak Bengali, the world's seventh most-spoken language, with almost no AI tutoring built for them.
~1.85M
Launch marketstudents sat Bangladesh's SSC and equivalent exams in 2026 — one predictable, deadline-driven cohort every year.
Why now
- Frontier models now reason, speak and read handwriting well enough in Bangla to tutor step by step.
- The cost of running a capable model has collapsed, so a full tutoring session can cost cents rather than dollars.
- Video-course platforms already trained families to pay for online learning; the next step is help that answers back.
- Curriculum changes leave existing content libraries out of date, opening a window for a new, AI-native library.
02
The AI case
What AI makes possible here, why the big labs won't absorb it, and whether the margins hold.
AI leverage — what is newly possible
- Explains any problem step by step in spoken Bangla, then checks understanding with a follow-up question.
- Reads a photo of handwritten working and points to the exact line where the student went wrong.
- Builds a per-topic mastery map from every answer and picks the next question to close the weakest gap.
- Writes the parent a plain-language weekly report — the job a human tutor rarely has time for.
Lab-proof moatsA reason competitors, including big AI companies, can't easily copy you: data, distribution, trust or local knowledge.
Curriculum graph
A topic-by-topic map of the national syllabus, past papers and marking schemes, vetted by teachers. Labs build general tutors, not this.
Student mastery data
Every answer improves the next recommendation. After a year, the model of how these students learn each topic is proprietary.
Parent trust and distribution
Schools, coaching teachers and parent groups are the channel. Trust is local and slow to copy.
Inference economicsThe cost of running AI models each time a customer uses the product. It decides whether an AI business makes money. (estimates)
- Unit
- Active student per month
- AI cost per unit
- ~$1.00
- Price per unit
- ~$4.00
- Gross margin
- ~70%
How the margin is protected
- →Route easy questions to a small model; reserve the large model for multi-step reasoning.
- →Cache explanations for the most-asked questions in each chapter.
- →Cap daily voice minutes per plan; offer unlimited text.
03
Market & expansion
Launch in Bangladesh. Then expand.
- SSC and HSC bring millions of students to fixed exam dates every year: a large, predictable, deadline-driven market.
- Parents decide, students use. Marketing must prove results to parents.
- Willingness to pay is proven by coaching spend; the question is price point, not whether to pay.
- Validate: which subjects drive coaching spend (math, physics and English are typical) and the monthly price ceiling.
Editorial scorecardPlankito's own judgement: the average of six 1–5 ratings (AI leverage, moat, competition gap, capital efficiency, timing, scalability). It is an opinion to compare ideas, not a forecast.
- AI leverage
- 5/5
- Moat durability
- 4/5
- Competition gap
- 3/5
- Capital efficiency
- 4/5
- Timing
- 5/5
- Scalability
- 5/5
Where it travels next
Bangla speakers worldwide
West Bengal and the diaspora share the language; the voice tutor works on day one.
South Asia
Exam-driven systems and heavy private tutoring in India, Pakistan, Nepal and Sri Lanka.
Africa & Southeast Asia
The curriculum graph and tutoring engine adapt to any national exam in its local language.
04
Competition & edge
Who already plays here, what they leave open, and what could go wrong.
Video course platforms (e.g. 10 Minute School, Shikho)
Recorded lessons, live classes and test prep.
GapOne-to-many content; they do not answer a student's specific question at 11pm.
Global AI tutors (e.g. Khanmigo)
AI tutoring built on English-language curricula.
GapNot mapped to local syllabi or exams, and weak in spoken Bangla.
General AI chat apps
Assistants students already try for homework.
GapNo curriculum alignment, no mastery tracking, no parent view — and they often just give the answer.
Where the edge comes from
- Voice-first in Bangla: works for students who type slowly or switch between Bangla and English.
- Photo-of-notebook grading matches how students actually study.
- The parent dashboard is what keeps parents paying month after month.
Risks and how to handle them
The AI gives a wrong answer and loses trust.
Ground answers in the vetted curriculum graph, show working step by step, and route flagged answers to teacher review.
Incumbent platforms bolt on an AI chatbot.
Compete on depth — grading, adaptivity and parent reporting — not on a chat box.
Inference cost grows faster than revenue.
Model routing, caching and per-plan usage caps; track AI cost per active student weekly.
05
Business model
Family subscription
Monthly plan priced below one coaching batch fee.
Exam-season passes
Short, high-intent plans before SSC and HSC.
Schools (B2B)
Teacher dashboards and homework grading for private schools.
06
Stack & data
Modern, proven tools a small team can run without a DevOps hire.
Web app (PWA)
Next.js 16 (App Router) + React 19
An installable PWA reaches every Android phone without app-store friction.
Styling
Tailwind CSS v4 + shadcn/ui
Clean, accessible components in days.
Tutor model
Claude API — a large model for tutoring, a small model for routing and quick checks
Strong reasoning and multilingual quality; tiered models keep cost per student low.
Speech
Speech-to-text and text-to-speech with measured Bangla accuracy
Voice is the core interface; choose the vendor by benchmark on real student audio.
Retrieval
PostgreSQL + pgvector
Curriculum chunks and embeddings live beside user data in one database.
ORM
Drizzle ORM
Type-safe SQL and simple migrations.
Auth
Better Auth (phone OTP, linked parent–child accounts)
Families share devices; linked accounts matter.
Evals
Answer-accuracy test set per chapter, run on every prompt or model change
Accuracy is the product; regressions must be caught before students see them.
Payments
bKash / SSLCommerz subscriptions
How families actually pay.
Hosting & ops
Vercel + Cloudflare R2 + Sentry + PostHog
Zero-ops deploys, cheap media storage, errors and learning analytics.
Data model7 core tables, for the engineer starting the first migration.
users
Students, parents, teachers.
- id
- phone
- role
- name
- class_level
- parent_id
topics
The curriculum graph.
- id
- subject
- class_level
- chapter
- title
- parent_topic_id
content_chunks
Vetted explanations for retrieval.
- id
- topic_id
- body
- embedding (vector)
- source
- reviewed_by
sessions
A tutoring conversation.
- id
- student_id
- topic_id
- started_at
- minutes
- ai_cost_usd
messages
Turns within a session.
- id
- session_id
- role
- text
- audio_url
- image_url
- model
mastery
Per-topic skill estimate.
- student_id
- topic_id
- score
- attempts
- updated_at
eval_runs
Accuracy checks before release.
- id
- model
- prompt_version
- chapter
- accuracy
- run_at
07
Costing
Directional monthly estimates in USD. Your numbers will differ — the structure is what matters.
| Item | MVP / monthWhat the product costs to run each month (hosting, AI models, tools) at MVP stage. | Scale / month |
|---|---|---|
| Hosting (Vercel Pro) | $20 | $300 |
| Database with pgvector | $25 | $250 |
| Model inferenceMain variable cost; capped per plan | $150 | $4,000 |
| Speech (STT/TTS) | $50 | $1,200 |
| Object storage (R2) | $5 | $60 |
| Content reviewers (part-time teachers) | $400 | $2,500 |
| Total | $650 | $8,310 |
One-time setup
08
MVP & roadmap
Shippable in about 6 weeks, then a path to production.
Build
- One subject (SSC math), one class level.
- Ask by text, voice or notebook photo; get a step-by-step Bangla explanation.
- Ten-question topic check that updates mastery.
- Weekly parent summary over WhatsApp.
- bKash monthly payment with a free trial.
Not yet
- All subjects
- Native apps
- School dashboards
- Live human tutors
Roadmap to production
Validate
Weeks 0–3- Tutor 20 students yourself with an AI model behind you on WhatsApp
- Measure questions per student per week
- Test price with 10 parents
MVP
Weeks 4–10- Ship the PWA for SSC math
- 100 paying families
- Answer accuracy above 95% on the eval set
Traction
Months 3–8- Add physics and English
- Exam-season pass
- Referral loop through schools
Production
Months 9–18- HSC coverage
- B2B school plan
- Second language market
09
Start this week
Your first steps. Tick them off as you go.
Metrics that prove it is working
- Weekly active students and questions per student
- Trial-to-paid conversion
- Topic mastery gain over four weeks
- AI cost per active student per month
10
Sources
Every figure on this page links to its source. Facts reviewed 14 October 2026. Costs and market notes without a citation are our estimates.
- 1.World Bank, UNESCO, UNICEF — 70% of 10-Year-Olds now in Learning Poverty (2022)
- 2.UNESCO — Global report on teachers: what you need to know (2024)
- 3.Dhaka Tribune (citing Ethnologue, 22nd ed.) — Chittagonian, Sylheti ranked among 100 most spoken (2020)
- 4.The Business Standard (BSS) — 18.5 lakh candidates to appear in SSC, equivalent exams (2026)