Government pilot · proposal stage
A measured AI learning pilot for Shahada and Taloda
Status, stated plainly: this is a proposal we are preparing for the Nandurbar Zilla Parishad Education Department. No department, school or official has approved, funded or joined this pilot yet. Every number below is a working draft that the Department would set.
What the pilot is for
Government school classrooms often hold children at very different levels, taught by one teacher. The pilot tests whether an AI tutor and teacher assistant, used with the teacher in charge, helps children learn more in Marathi and English, and whether it is practical to run at district scale. The Department decides the problem the pilot should address; we bring the platform, the training and the measurement.
Proposed scope (working draft)
- Where: selected government and Zilla Parishad schools in Shahada and Taloda blocks, Nandurbar district.
- How long: six months in classrooms, after set-up and approvals.
- How many: about 10 to 25 schools, 1,000 to 3,000 students and 50 to 150 teachers. The final sample is agreed with the Department after school selection.
- Languages: Marathi and English, voice-first for younger children. Hindi only if the Department asks for it.
- What is used: personalised practice, the AI tutor within teacher-set limits, a teacher assistant for practice sheets and next steps, assessment, AI literacy lessons, and teacher and district dashboards.
- Connectivity: designed for low bandwidth. Offline use is planned; how well it works is part of what the pilot tests.
How it runs
- Approvals first. The Department names a nodal officer and selects schools. Data rules are agreed in writing before any student data is collected.
- Measure first. A baseline assessment in pilot schools and in comparison schools, before anything changes.
- Train teachers. Short, practical sessions for teachers and head teachers.
- Use in class. The platform is used alongside normal teaching, with the teacher deciding when and how.
- Check in monthly. Usage, problems and fixes are reviewed with the Department every month.
- Measure again and report. An endline assessment, compared against the baseline, and a public method. A result that shows no difference is reported as one.
What we will measure
- Learning: competency by grade and skill, baseline against endline.
- AI literacy: understanding of what AI is, when it is wrong, and how to use it safely.
- Teacher adoption: how often teachers use it and for what.
- Student participation and, where the Department authorises it, attendance trends.
- Equity: participation by gender, school and location.
- Reliability: availability, offline sync and support response time.
- Value to the Department: reporting time saved and visibility of learning gaps.
Targets are set only after the baseline. We will not publish illustrative numbers as results.
Children's safety and data
Teachers stay in control. The tutor removes phone numbers, email addresses and ID numbers a child types before anything reaches the AI model (it does not detect names, so children are told not to type them), stops and points a child to CHILDLINE 1098 if they say something that suggests they are unsafe, and flags that session for an adult. Children in Classes 1 to 5 use it only with an adult present. Pilot data is kept separate from development data, minimised, and handled under the Digital Personal Data Protection Act, 2023, with the Department's written authorisation. More on the safety section of our home page.
What we ask of the Department
- A short meeting to agree the problem the pilot should address.
- A nodal officer and a process for selecting pilot and comparison schools.
- Written data rules: what may be collected, who may see it, and how long it is kept.
- Time for teacher training within the school calendar.
Contact
Download the 2-page pilot concept note (PDF, 76 KB).
Write to [email protected] for the detailed proposal, budget and data rules.