Last week, we kicked off AI Cohort 2.0. It made me pause and think about the journey with Cohort 1.0 — especially my time working with Avanti Fellows.
Earlier this year, we wrapped up Cohort 1.0, and Avanti was one of three NGOs — alongside Avni and Inqui-Lab Foundation — selected to continue into Cohort 2.0, along with Reap Benefit, Madhi, Glific and U&I Trust
Setting the Scene
The AI Cohort Program is run by Project Tech4Dev and The Agency Fund– a hands-on program that helps NGOs design, build, and scale responsible AI solutions through dedicated mentoring, prototyping support, and peer learning over 4 months. Working at Tech4Dev as part of the Kaapi team (an AI platform for responsible AI in the social sector), I was assigned to mentor Avanti Fellows for this cohort.
For context: Avanti Fellows is a non-profit that provides free, high-quality science and maths education to students from low-income backgrounds. They work with students in grades 9–12, helping them prepare for JEE and NEET. They run centres in government schools, live online classes, and self-paced programs — reaching over 100,000 students.
The Journey: Four Months, Four Phases
To best describe the journey, it was like setting off on a small raft in the open sea. There were pleasant days, stormy nights, and moments where you were just figuring out which way the wind was blowing. But at the end, we reached our destination and look back and realise how far we’ve come.
We were figuring things out as we went – doing this for the first time – and Avanti was juggling between keeping their program running smoothly on the ground while carving out time to invest in the future by experimenting with AI.
Looking back, the four months can be put into four phases:
Phase 1: Exploration and Scoping
We spent the initial weeks specing out usecases — Engagement (replicating high-touch mentoring for 80,000+ low-touch students via AI), Recommendation (personalised test-prep strategies after each test), and User Segmentation (combining qualitative and quantitative data to create student personas) – using the Knowledge Graph + LLM approach. Really enjoyed working with Poojita, Deepansh and Dhyanesh to flesh these out. It helped me understand more about Avanti Fellows, and hopefully, each of these usecases will eventually find its way into their program in one form or another
Phase 2: Building Momentum
We then decided to focus on “AI Mentorship Notes” for this cohort. AI Mentorship Notes is an AI-generated summary to help teachers deliver personalised test-prep guidance. Teachers were spending a lot of effort manually comparing test reports and cross-referencing JEE cutoffs and handwritten notes for each of their 10-15 mentees every test cycle.
The hypothesis: AI-generated performance summaries can enable personalised mentorship with far less prep time.
Loved seeing Deepansh build out a quick prototype to get things started
Generated AI Mentor Notes
Phase 3: The Quiet Middle
Testing on the ground with staff members while keeping the program alive is crucial as the JEE and NEET approaching. The team ran a thorough feedback loop (2 mentors + 4 expert reviewers) evaluating AI notes on accuracy, clarity, hallucinations, and more.
Incorporating the feedback, the Avanti team quickly iterated on the prompt.
Phase 4: The Final Sprint
Consolidating everything for the closing ceremony. In the final presentation, they told the story through a real student, Khushboo, a first-generation college aspirant from Punjab, and her mentor, Bhumika Ma’am, showing the manual 5-step process teachers go through and how the AI tool transforms it. A great presentation by Dhyanesh and Surya, with Poojita being the shepherd who kept everything together.
What Worked Well
- Dedicated communication channel: A Discord group for quick exchange of messages and coordinating meetings helped a lot in moving things forward
- Weekly sync-ups: Regular check-ins to track progress and course-correct. Communication really is the key — not just the big strategy calls, but the small “here’s where we are, here’s where we’re stuck” conversations.
- Documenting everything: From the first 3–4 use cases to meeting notes and action items, a single Google Doc kept things from slipping through the cracks and helped everyone quickly get back on the same page and plan the next steps.
- Monthly feedback and check-ins: These helped realign on the end goal whenever things started drifting. In a 4-month engagement, it’s easy to lose sight of the big picture.
- Experimenting fast: Getting things on the ground rather than chasing perfection. The team tested the AI notes with teachers, gathered feedback, and iterated.
- Not jumping to solutions: We didn’t need to have all the answers from the start. Admitting “we don’t know” and figuring things out together with the NGO, keeping their usecase in mind, was far more productive than pretending to be experts.
- Pure honesty: If things were falling behind, we communicated that. And still maintained a healthy relationship — because at the end of the day, we all want the same thing: to help serve the community better and more efficiently.
Things still on my mind
Cross-mentor collaboration: I know external mentors have limited bandwidth, but better coordination could help us support each other—bringing fresh eyes to the work different NGOs are doing will help in building shared wisdom and a stronger community. I hope we figure this out in Cohort 2.0.
At last, I’m just grateful to have been part of this journey. Through Cohort 1.0, we at Tech4Dev gained clarity on how to run AI cohorts—lessons we’re already applying to the next round. I also learned the nuances of mentoring NGOs: when to push, when to listen, and when to step back so teams can own their process.
Do checkout the blog by Ashana on our learnings from AI Cohort 1.0
With Cohort 2.0 now underway, I’m excited to be working with old friends at Reap Benefit alongside Jerome while staying in touch with Avanti Fellows to see what they accomplish this time around.
