Who you work with#
Project SparkL, described in Brief 06. It works with underprivileged children, many of whom have run away from home, who are placed in a school grade because of their age rather than what they can do. SparkL’s facilitators spend the hours after school closing that gap.
A detail that matters: a child’s grade hides the fact that their reading comprehension, their writing and their reasoning can all be at different levels.
The situation#
SparkL has the design, the curriculum and the facilitators. All of it is offline and manual.
Every week, every facilitator travels to one location and spends a full day being trained. The founders spend that day with them — teaching, going through feedback, and analysing what happened in the sessions. That day is where the quality comes from.
What is going wrong#
The facilitator is the product. Quality control at the facilitator level is the single thing that makes SparkL work, and it currently requires the founders in a room once a week.
So growth costs founder-days, and there are only so many of those. Seven centres is roughly where that ceiling sits.
Why it is not fixed yet#
Because recorded training does not build this skill. Teaching a child who has run away from home, who is nine and cannot count, and who may not want to be there, is a practised interaction. You cannot learn it from a video any more than you can learn to swim from one.
There is a second reason people miss. That weekly day runs in both directions. Training goes down to the facilitators, and feedback and analysis come back up to the founders. A platform that only pushes content downward removes half of what the day does.
Skills and Expertise#
- Branch — CS, IT. The most demanding language-model problem in this pool, and not a wrapper exercise.
- Technical skills — Controlling and constraining model behaviour, evaluation design for open-ended systems, conversation and dialogue design, state modelling, offline and low-bandwidth delivery, Kannada language handling.
- Other skills — Understanding pedagogy before automating it, designing a training experience, measuring whether a person actually got better, working with vulnerable-group safeguarding.