Even though this is an “AI” cohort, we’re starting out by building a rule-based recommendation engine where we know the key parameters and have identified their thresholds. Our primary goal with AI Cohort is to:
Evolve from
“a single chapter recommendation based on one test report”
to
“comprehensive actionables based on all tests taken – covering everything students should focus on and avoid”
While this is our immediate goal, we’re also thinking long-term about leveraging AI to identify underlying weak areas based on attempt patterns, beyond our current tagging system.
Introduction Session
Starting with an intro session (facilitated by a structured template from T4D beforehand!) was great as we immediately knew what others were doing and how we could collaborate instead of awkward networking later. Right away we started connecting others’ work not just to our specific use case for the AI cohort but to other product verticals within Avanti Fellows. This facilitated much deeper discussions throughout the two days.
DFL Interactive Session
The DFL workshop was the best part. It wasn’t just engaging but it also made us think about ensuring that while our end users use our tools, we’re also building capacity within them. In retrospect, this applies to all our tech solutions and not just limited to AI-driven tools.
This is interesting for our recommendation product because of how this problem statement came about in the first place:
Recommendation Engine Journey
(highlighted in blue are quotes from student feedback)
Through this journey, “capacity building” took a backseat as we are constantly trying to figure out the most direct instructions possible. While our rubric for the next version would ensure students can meet baseline requirements by exactly following this simple guidance as it is, there’s actually room to help motivated students create their own study plan/ strategies based on various personal factors only they know.
Learning from other organisations
Simple Education’s teacher assistant bot on WhatsApp caught our attention. With our extensively detailed teacher training modules, we can see how providing teachers with the option to reflect with an AI bot about classroom dynamics in a leisurely and unforced way could drive best practices, especially our focus on equal engagement of girls in classrooms – even if it just helps them think about it more, that’s a big win.
Quest Alliance’s approach of identifying thirty-three macro behaviours and grouping them into five personas and creating solutions for each made us think beyond just “what” we recommend to “how” we recommend, personalised to each student.
💡Next step for us: We are now considering creating user personas for our context to determine the recommendation language that helps students understand, retain and follow guidance.
Just when we felt like we had enough new concepts to explore, Inqui-Lab’s roadmap took personalisation to an entirely different level – creating personas not just through text responses but by capturing emotions that can’t be obtained through text alone. They’re in the ideating phase and looking into video/ audio cues to gather deeper user understanding for profiling.
💡This made us wonder: Which use cases would require us in Avanti to consider students/ teachers not just as a collection of messages to analyse and respond to, but as people with reactions and emotions that are actually crucial for our responses? Maybe the in-house counseling module, maybe a chatbot for the asynchronous students for guidance on recommendations? I honestly hadn’t thought about this until Ajay elaborated on the need to understand users beyond their written inputs.
Collaboration Opportunities
Avanti Fellows covers almost the entire student journey (Grade 11 to college and beyond) → from selection to career guidance, teaching, testing, evaluating, counseling, college selection, scholarships and school-to-work. These are different verticals within the organisation, in various degrees of scale and depth. Many non-profit organisations specialise deeply in some of these areas where we can collaborate for mutual learning since while we have depth in testprep, we are still developing our other verticals and have a lot to learn from organizations who have already been working deeply for a while in those spaces.
Shoutout to the T4D team for tabling a small discussion with the folks from a few non profits who are working in the education / training / upskilling space. Those conversations were very fun and gave us a lot to think about. The work Quest Alliance is doing can be very useful for another vertical we’re working on, so it was great to connect with them!
Not all our verticals are tech-driven or tech-facilitated (yet). Our product-tech team is currently understanding the various solutions they’ve developed and their pain points. This gives us ground to:
- Identify where AI can simplify internal processes
- Collaborate with organizations solving similar use cases and learn from them
What’s Next
This workshop shifted our thinking beyond just improving our recommendation engine. We’re now considering capacity building, personalisation approaches and collaboration opportunities. The conversation around user personas from both Quest Alliance and Inqui-Lab opened up an entire dimension of not just what we tell students but how we tell them based on who they are.
Written by Poojita Srinivasan & Deepansh Mathur from Avanti Fellows
