A state's first classroom-monitoring system, from an offline prototype to the state's own app

Chapter 1 · The problem

A state retrained its early-grade teachers, but nobody above a school could see a single lesson

In 2021 and 2022, one Indian state set out to fix early reading and math. It retrained about 50,000 teachers and gave them new handbooks and workbooks. But when someone visited a school, the only checklist was a paper one about things like midday meals, and nobody tracked it.

I worked on the mission for Central Square Foundation (CSF), the nonprofit that supported the state. The mission had five parts: learning goals set against a baseline, new teaching materials, teacher training, mentoring in the classroom and monitoring. The last two depended on something the state did not have. Nobody could say whether a teacher was using the new lesson steps, whether a mentor had visited her, or what the children in her class could do.

This state was not unusual. In most states CSF reviewed, school monitoring counted buildings, attendance and enrollment. A visit happens, or does not, and nothing about teaching reaches anyone who can act on it.

Monitoring tools in most states counted inputs, not teaching

CSF's 2023 article on the program puts the cause plainly. Inspection visits existed, but they were not structured and did not look at academic problems. The state also lacked the people and resources to run monitoring well.

Sources: CSF's review of monitoring tools across five states (funder presentation, c. 2023); CSF's 2023 article on the program; the mission design, 2021 to 2022.

Chapter 2 · The system

We designed the mentor role, the visit and the training first, and left the tool until last

The department and CSF planned monitoring and mentoring in four parts: process, people, capacity and technology. I pushed for the tool to come last. An app can record a visit, but it cannot decide who should visit, what they should look for or what they say afterwards.

Headmasters became mentors, because teachers respect them

The obvious choice was the cluster resource persons already on the payroll. I argued against it. They were on contracts, teachers saw them as data collectors, and they had little time. We chose senior cluster headmasters instead, with the senior-most one in each sub-district as its nodal officer, in a cadre of about 10,000 mentors.

Every visit follows the same steps, and teachers are protected from too many

The protocol also caps visits. No teacher should get more than one visit a month from all officials combined, and no school more than one a week. Each cluster headmaster visits every school in the cluster once a month. Each nodal officer visits at least 12 different schools a month and covers the whole sub-district in about three months.

Sources: CSF's 2023 article on the program; the 2024-25 visit diary (visit protocol, school-visit checklist); the mentor decision and cadre size, 2022.

Chapter 3 · Offline first

Mentors were all in training for a few weeks, so I built an offline tool for that window

The state's app was due in January 2023. The mentors' training ran in August and September 2022, the one time all 10,000 would be in a room together. If their first four months of visits went unrecorded, I expected the habit to fade before the app arrived.

I already knew a tool that worked without a signal

Earlier that year, in March and April 2022, the state ran a sample baseline test of Grade 3 children. CSF had done this in another state on pen and paper. I compared assessment tools and chose Tangerine, an open-source offline assessment app made by RTI International. I convinced CSF to pay for it, about Rs 5 lakh a year (about $5,700 at Rs 88 to the dollar), so it could be reused across states. Then I learned its authoring language and built the tests myself, including a timed reading test that works out fluency on its own.

The survey tested about 6,000 Grade 3 children in each subject, one to one, through over a thousand trained field investigators. Printing costs fell to almost zero.

I re-authored it for a different job in three weeks

For the mentors, I re-authored Tangerine as an offline-first visit tool with two parts: a classroom observation against the lesson's expected steps, and a spot test of five children chosen at random. It was offline-first because weak networks were a big reason earlier government apps had failed in schools. I tested it with about 20 users, ran a pilot, then rolled it out to the mentors in their training sessions.

The hard part was owning it

CSF had never owned software. If it went down, or data was lost, the failure was ours. I had to convince my manager, then the organization, then the state's education secretary and the commissioner.

CSF's 2023 article describes the approach as planning carefully, starting small with a minimum viable product, then growing it quickly.

The tool ran until the state app replaced it in December 2022.

Sources: CSF's 2022 article on the baseline assessment, for the sample and the investigators; CSF's 2023 article on the program; the build, tests and approvals, Aug to Sep 2022.

Chapter 4 · What the data showed

In four months, mentors logged about 50,000 observations and 150,000 student assessments, and 38% of observed teachers improved

In the first four months, mentors logged about 50,000 classroom observations and 150,000 student assessments, and reached about 15,600 schools. Each visit now left a record that someone above the school could read.

~50,000

classroom observations in the first four months

~28,600

teachers visited at least once in the first four months

150,000

student assessments in the same months

38%

of observed teachers improved their practice, first to last visit

The dashboard was free, and it counted the lesson plan's own steps

Data helps only if someone with power looks at it, and there was no budget. So I built the dashboard on free tools. A SQL database on Google Cloud joined the tool's exports with the state's student, teacher and school IDs. A Python script refreshed it on its own. Looker Studio showed it.

The hardest part was the counting rule. A regional-language lesson had six expected steps and an English lesson eight, with variations by type of class. So "followed the lesson plan" had to mean something different for each, or the dashboard would mark good lessons as failures.

Measured The dashboard tracks the share of steps followed every month. 38% of observed teachers improved their practice between first and last visit, and the teaching practice score improved by 30 percentage points.

Sources: Program records, for the 50,000 observation total; CSF's 2023 article on the program, for every four-month count; the dashboard build and its counting rules, Sep 2022 onward; my arithmetic as labeled.

Chapter 5 · The state app

I chose the state's own vendor and led its engineers myself, so the state could keep the app running

The offline tool was a stopgap. From June 2022, in parallel, I wrote the vision for one app for the whole mission: textbook and workbook delivery, teacher attendance, training, classroom observation and student assessment, with one dashboard on top.

To my knowledge, as of 2022, none of the states CSF worked with had one app covering all of it.

I wrote the vision, the roadmap, the user stories and the acceptance tests, and worked with the engineers module by module. There was no designer on either side, so I designed the screens in Figma and used clickable prototypes to get the department to agree on the app before any code was written.

What mentors struggled with on the offline tool became the app's requirements

The app gives each role its own job

The article groups the app into four parts: student learning, the teaching-learning process, governance (visits and review meetings) and critical enablers such as textbooks and libraries. Each role sees its own part.

The dashboards moved to Power BI with row-level security, so each official sees only their own district, sub-district or school. I designed the screens, the user flows and the indicators, set up the data pipeline from the app, and tested the data.

Sized for every teacher, and still running

I built it for about 150,000 users and sized the servers from expected weekly use, because every teacher in the state would run assessments in it, not only the 50,000 in the mission. What broke in the stopgap became its requirements. The app was tested with users again and launched in December 2022, 30 days ahead of its January 2023 deadline, and mentors moved over from the offline tool. It was adopted as a model by 11 states reaching 20 million children.

Live The app's store listing shows more than 100,000 installs. The vendor and the department have added modules since I moved on.

Sources: The vendor decision, design, sizing and launch, Jun to Dec 2022; CSF's 2023 article on the program; the app's store listing, accessed 1 Nov 2026.

Chapter 6 · Adoption

Use was tied to routines mentors already had, the visit checklist and the monthly review, as one district's records show

A year earlier, during school closures, I had rolled out a reading app across the same state's schools. Daily users peaked at about 10,000 after the orientation, then fell to about 2,000, because the message did not travel down to teachers and parents. I did not want the observation app to follow the same curve.

Entering the data became a step in the visit

So the app was not launched as one more tool. It was launched as the single place for the whole mission, with launch messages planned from day one. Entering the observation became the last line of the mentor's own visit checklist (Exhibit 3). And the numbers it produced became the agenda of the meetings where officials are reviewed.

The records below come from one district my team supported, in its 2024-25 visit diary.

Mentors pair a paper plan with the app entry

In the 2024-25 diary, mentors plan visits and take notes on a printed sheet, and the sheet asks them to tick "Observation entered in App?". Paper works without a signal and feels familiar. It also means each visit is entered twice, which Chapter 7 takes up.

Training was measured too, in one district

Measured, 29 mentors This measures what mentors knew after training, such as the state's approach to teaching math with objects, then pictures, then numbers.

Sources: A deck on the reading-app rollout (2021); the 2024-25 visit diary (review checklists, data sheets, observation sheet); the September 2024 training pre- and post-test, aggregates only.

Chapter 7 · What comes next

Teaching practice improved, so the next tests are attribution and feedback

Counting visits was the first job, and it worked: 38% of observed teachers improved their practice, and the teaching practice score rose 30 percentage points. The next job is to measure how much of that change the visits account for, and how good the mentor's feedback is.

Give mentors their own reason to enter data

Monthly review meetings got data flowing across a state in four months, because mentors' numbers came up in front of their seniors. The next step is to make the data useful to mentors themselves: a view of each teacher's lessons over time that helps them plan the next visit and the next piece of feedback.

Keep the visit supportive as the data travels upward

The same drill-down that lets an official find a struggling sub-district can show one teacher's lessons. The protocol already tells mentors to present each visit as support, and caps how often a teacher is visited. Next, I would ask a sample of teachers how visits feel, and track that next to the visit count.

One entry per visit

I would retire the paper visit diary once the app is in use, so mentors record each visit once and nobody has to check that the two match.

Sources: CSF's 2023 article on the program; the 2024-25 visit diary; the app's store listing, accessed 1 Nov 2026.

Reflections

A stopgap built by hand wrote the state app's requirements