evidoria

← Back to browse

Good practice Imported

Project Soro — Tajikistan's National AI Curriculum and Tajik-Language Chatbot for 2 Million Students

Tajikistan · Dushanbe · See the Tajikistan profile · See the Dushanbe profile

Evidence: Descriptive / self-reported Top 28% 60/100 · Ask Evidence Copilot about this practice

Tajikistan's AI Council, UNICEF and zypl.ai launched Project Soro in Oct 2025 to make 'Intro to AI' mandatory in all schools by 2027, deploying the Tajik-language SoroLLM chatbot; a pilot reached 30 Dushanbe schools (770 teachers, 2,000+ students) before scaling to 100 schools.

30
Schools in initial Dushanbe pilot
770
Teachers reached in pilot
2,000+
Students reached in pilot
100
Schools reached at time of reporting
4,000 schools / 2 million students
National rollout target by 2027 (by 2027)
Project Soro — Tajikistan's National AI Curriculum and Tajik-Language Chatbot for 2 Million Students

Details

Maturity
Pilot
Promoter
AI Council of Tajikistan / UNICEF Tajikistan / zypl.ai
Period
Launched Oct 2025; mandatory subject and full national rollout targeted for 2027
Keywords
AI literacy, K-12 curriculum, Tajik-language LLM, teacher training, inclusive education

Context

Project Soro is a joint initiative of Tajikistan's Artificial Intelligence Council, UNICEF Tajikistan and the Tajik AI startup zypl.ai, launched in Dushanbe in October 2025 to integrate AI into the national school system, with a mandatory 'Introduction to Artificial Intelligence' subject planned for all of Tajikistan's roughly 4,000 schools by 2027.

Objectives

The plan aims to reach an estimated 2 million students nationwide and train up to 2,000 AI specialists, with government and partner projections putting the initiative's economic value at around $150 million.

Activities

The technical core is SoroLLM, a family of lightweight Tajik-language conversational models built through continual pretraining on a curated 1.9-billion-token Tajik corpus, designed to run under the country's compute and connectivity constraints and integrated into the existing maktabmobile.tj and eDonish school platforms. Roles are split explicitly: the AI Council coordinates and sets standards, zypl.ai builds and deploys SoroLLM, and UNICEF Tajikistan is responsible for ethical compliance and child-rights safeguards.

Results

An initial pilot in around 30 schools near Dushanbe reached 770 teachers and just over 2,000 students; by the time of reporting the model was being piloted in 100 schools nationwide, still well short of the 4,000-school, 2-million-student target for 2027. No independent learning-outcome evaluation has been published.

Implementation
Indicative cost
High (€500k–€5M)
Time to results
Long (> 3 years)
Staffing & skills
AI Council of Tajikistan coordinates and sets standards, zypl.ai builds and deploys the SoroLLM models, UNICEF Tajikistan is named as responsible for ethical compliance and child-rights safeguards

Conditions for success

  • SoroLLM designed as a lightweight model to run under tight national compute and connectivity constraints
  • Integration into existing maktabmobile.tj and eDonish school platforms rather than a standalone system

Common failure modes

  • Scale-up from around 100 to 4,000 schools remains almost entirely ahead and unproven
  • No independent learning-outcome evaluation published to date

Where it fits

Governance type
national government with multi-partner (UN/startup) delivery
Scale
national
Income level
low-income

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful