Radics

A light-weight, AI-chat driven learning data management system for under resourced schools.

Receiving consistent and personalized support from Parents, Teachers, and School Leaders is the cornerstone of imparting foundational skills to students in the K-8 group. Regular and reliable diagnosis of learning progress is a key enabler for this support. However, ability to frequently collect and utilize learning outcomes data remains limited to elite and mid-income schools. We want to change this with Radics.

Radics is a light-weight, AI-chat driven learning data management system for under-resourced schools. It enables cost-effective collection of high-quality learning outcomes data in both online and offline settings via WhatsApp and OCR-enabled worksheets. It then transforms the raw student responses into byte-sized insights for Teachers, Parents and School Leaders, shared via chat messages and interactive dashboards. These insights are further enriched with evidence-based micro-actions that these stakeholders can take themselves or nudge others to take in order to provide personalized and contextual support to students.

We incorporate research at four levels.

1. Teaching at the Right Level (TaRL) - Evidence from India and Africa suggests that reorienting instruction to meet the student’s learning levels, irrespective of the mode of instruction, can have significant improvement in learning outcomes. Our data analytics engine is capable of culling out insights at individual student level [J-PAL and Pratham]

2. Sharing data with Parents - A series of experiments across the developing and developed countries prove that providing parents with clear, actionable guidance on how to be more involved in their child’s academics has improved learning outcomes for their children. Timing and design of texts plays an important role, and we consider this in the design of our nudge messaging system [JPAL]

3. Framework for Data use by School Leaders - school leaders can drive visible results in student outcomes by incorporating data use at two levels - a) Macro - to connect student test results to system outcomes such as curriculum design, b) Micro - to drive daily teacher and parent behaviour change. We design our School Leader interfaces knowing they are the bridge between the Macro and Micro changes in schools. [Alex J Bowers et al - Teachers College, Columbia University]

4. Contextual interfaces for EdTech solutions - a rapid evaluation of 12 edtech interventions by CSF revealed that most ed-tech solutions are not well-designed for low-income students and thus fail to live up to their promise because of their inability to contextualize. The key factors to consider for contextualization are prioritizing vernacular mediums, and using simple to use interfaces - two core considerations in our approach.[CSF]

Radics flips the learning data management problem from being an organizational concern to a consumer issue. Most learning data management solutions start by defining what school administration needs as an organization as opposed to what the individual stakeholders’ pain points are. Meeting students, teachers, parents and school leaders where they are can have a significant impact on how they interact with technology and use data. Adhering to this principle becomes further challenging for low-resource settings that introduce additional constraints related to language, tech literacy and online-offline synchronization. We innovate on three fronts to address our specific challenges and practice a consumer mindset.

One, users get to interact with our system through a conversational interface that’s personalized for their specific needs. Two, we complement online-chat with low-inference, tamper-proof data collection strategies such as OCR-enabled worksheets, digitized using AI image processing. And finally, we leverage machine learning to break raw datasets into actionable insights that have a known and proven impact on student outcomes. 

We believe that our approach could potentially pave the way for incumbents and newcomers alike to focus on building contextual technology solutions, and leverage AI/ML to tackle other unresolved barriers to imparting quality education in low-resource settings.

Data from a growing body of TaRL (Teaching at the right level) interventions prove that knowing a student’s right learning level, and then tweaking instruction to match those levels can have a significant impact on student outcomes. Variance in student outcomes as the delivery modes change from computer-based to teacher or volunteer-led only has marginal effects on student outcomes. This insight is at the heart of our theory of change. We intend to build a system where the key providers of education i.e. School Leaders, Teachers and Parents are not a mere agent in a simplified data collection process, but are actively utilizing data to modify their behaviour towards better student outcomes. This makes our approach a human-machine collaboration to meet student’s needs, rather than an attempt to completely bypass the human support layer. 

We realize our hypothesis by working closely with Teachers and School Leaders without introducing new technology barriers. Data from our pilot in Maharashtra reveals that access points like WhatsApp chat can lead to a significant increase in promotion of student data collection and usage practices, primarily led by Teachers, and promoted by School Leaders and Parents. Over the last 3 months, we’ve generated more than half a million student test responses from just over 400 schools, using an online test taking approach. We believe that by adding a set of offline data collection tools we can improve our student reach by a factor of 2 to 3.

In the long-term, we want to help under-resourced schools practice learning data management at par with an elite school, without having to invest in expensive ed-tech tools. In some ways, this transition will be similar to the proliferation of mobile data in developing nations where a large mass of citizens adopted 4G/5G, skipping the broadband internet.

By the end of 2021 we want to bring Radics to 5000 under-resourced primary and upper-primary schools, reducing barriers to good quality student data collection and data-driven decision making practices, and enabling personalized student support at scale. With-in these 5000 schools, we will ensure at-least 75% student and parent outreach via both online and offline data collection methods, and 100% adoption among Teachers and School Leaders.

We will primarily achieve these targets by working with nonprofits and district education departments, with a part of the growth (~20%) coming from our direct to school outreach efforts.

By the end of 5 years, we intend to be in a position where we could lead cross-country adoption of Radics as a globally recognized solution for learning and school operations management. We’ll achieve this by building strategic partnerships with leading NGOs, Government Bodies and large scale education reform sponsors such as the World Bank. Additionally, we’d like to experiment with an open sourced approach to further enable the adoption of our solution. By this point, we imagine Radics to be impacting the learning outcomes for 10 million+ students from 100K schools.

We see two technical challenges in the immediate term.

First is the access to statistically tested question sets and student surveys that’ll enable reliable comparison of data across student groups, classes, schools and the cluster of schools. Market based solutions exist for this but none offers the question bank as a service, and all charge a premium through a per-student or per-test business model. 

Second, for our actionable insights recommendation engine, establishing the links between the raw data and the contextual nudges for Teachers and Parents will require collection of high quality training datasets before the actionable insights are practically useful. 

In the long-run, we see sustainability as a challenge in a hyper-competitive and cut-throat ed-tech industry. Our focus on meeting the specific demands of low-resourced schools will further aggravate this challenge.

Statistically validated question sets - The only way a lean approach like ours can mitigate this challenge is by crowdsourcing questions from Teachers and NGOs, leveraging libraries like Open IRT to statistically validate the question items, and then making the question bank available as a free service usable via APIs to generate a positive feedback loop. This may upset the incumbents but will help create an open-source, plug-and-play alternative to a closed, proprietary and high-cost monopoly.

Nudge recommendation model - we’ll start by designing a small set of nudges representative of a part of our target audience e.g. students from a  specific age group, or a specific set of student competencies. We’ll use this sample to prove the effectiveness of the nudge recommendation system before covering all user bases. 

Sustainability - we’ll build organizational capacity to win long-term strategic partnerships with leading NGOs, Governments and Funders, as opposed to building transactional relationships as a service provider. To balance the high acquisition costs associated with long-term partnerships, we’ll launch a direct-to-school version of our solution, offering unbundled services such as admissions management to middle-income schools that have the capacity to pay.