Radics

A CASEL study confirms that targeted socio-emotional learning (SEL) interventions increase the chances of student success by 11 percentile points, compared to students who do not receive such interventions. More than 50% of 9-14 years old students from low-middle income countries currently do not get trained on essential socio-emotional learning skills, and without personalized interventions, 43% of these students will likely drop out of school or enter college with sub-optimal life skills [World Bank, 2022].

Designing targeted SEL interventions demands assessments innovation that can offer evidence-based and non-intrusive approaches to profile young students on all important SEL components. Self-regulation, one such critical SEL component, is evidenced to explain important transferable dispositions in students such as reflective thinking, goal orientation, growth mindset, and intrinsic motivation - skills and behaviors essential for learners from all age groups [Zimmermann, Pintrich]. Esp. for learners facing inequity at both school and home, self-regulation can be of special advantage as building this skill will help them better manage their surroundings and its effect on their well-being [Boekaerts et. al.].

Radics is a collaborative learning measurement and improvement platform that helps educators model the dynamic and sequential nature of student self-regulated learning skills, esp. metacognitive performance, as students go through a variety of learning experiences. The key objective of these decision models is to expose the causes of sustained low self-regulation and enable one-on-one and cohort-based support led by teachers, parents and life coaches. 

Radics offers contextual, valid and reliable measurement tools to track student metacognition and self-regulation skills - both digitally (web and chat) and through paper-based assessments, and distills currently known self-regulation intervention strategies and approaches into student-friendly narratives using AI-enabled data analytics. These narratives help students meaningfully engage with their self-regulation capabilities via three lenses:

• ? I know: a descriptive evaluation of aggregate student performance on self-regulation and metacognitive performance assessments
• ? I wonder: a set of question prompts that push students to critically evaluate their skills and draw inferences based on their lived-reality
• ? I do: micro-actions vetted by SEL experts and based on existing SEL intervention literature that students can adopt in daily life to build self-regulation skills

To enable a developmental approach for students to chart their own journey to become reflective learners, Radics also offers an qualitative view of student metacognitive skills using Perkins' metacognitive learner framework as follows:

Students from our target contexts (India and Kenya) have a daunting challenge of maintaining self-regulation through a set of macro (e.g. long-term goals), and micro (e.g. daily lessons) learning engagements that are not optimised for critical and reflective thinking behaviour. Modern self-regulated learning research proves that these macro and micro learning contexts, when fortified with specific self-regulated learning strategies, lead to superior student performance on a variety of academic and non-academic outcomes, effectively helping students build long-term, transferrable life skills. 

Understanding the dynamic and sequential interrelationships between these macro and micro learning contexts through multimodal student assessments is at the cutting edge of modern education research, but leaners from low-resourced contexts are still not a focus of this work – with Radics we want to change that.

We work with teachers, life coaches, local education non-profits, and school district administrators across India and Kenya to deliver cutting edge innovations in multimodal assessments of socio-emotional learning skills. Further, we enable action-oriented SEL assessments by deepening our engagement with the proximate child support providers through tight-knit communities using online channels, meaning making workshops and design hackathons.

We build and nurture stakeholder relationships through two pathways – providers and enablers. 

At the provider level, while most edtech solutions focus on scale by reaching students directly, we engage with teachers and parents in creating conducive learning environments. At home, this means routing learning tasks and assessments via parents and fortifying student responses with parental feedback. At school, teachers are provided lesson support to gauge the student engagement and joy in classroom, along with timely, actionable insights into student performance.

At the enablers level, a shared practice of socio-emotional learning is built through cross-org collaboratives where issues such as access despite digital divide, child rights and data privacy are at the heart of the conversation. A key outcome of our multi-stakeholder approach is an open assessment architecture for socio-emotional learning where our content and key research findings are open sourced for wider, sectoral benefits.

We co-design our solutions with our end beneficiaries, primarily students and teachers, by engaging them in focus group discussions, cognitive interviews and workshops. Over the last 18 months, these engagements have helped us ensure language and cultural fit for our SEL measurement devices, and establish the utility of our data analytics to drive corrective actions. While teacher workshops and hackathons gave us a platform to collaboratively design interventions with teachers and prioritize key product features such as short-cycle student feedback.

The rise of modern AI tools (e.g. ChatGPT, MidJourney, Bard) promises to answer some of the long-standing education concerns such as the 2-sigma problem (i.e. cost effectively scaling high quality 1-1 tutoring) posited by Benjamin Bloom in 80s. At the same time, the race to dominate the lucrative AI market has set things in motion that may lead to real-life catastrophes if not controlled. As a startup that finds itself in the middle of this major inflection point, faced with a choice whether to fully embrace the technology or completely oppose it, we sincerely hope that being part of a diverse ecosystem like MIT Solve will help us build a critically constructive  viewpoint towards artificial general intelligence and its various applications in education domain.

As we re-shape our approach to responsibly leverage AI for education, we will seek expert guidance on two important fronts - strategic advisory and research collaborations. Through the MIT Solve platform we are seeking support to build:

• an advisory board that aligns with our mission of evidence-backed SEL assessments and can guide our transition from pilot stage to a scaled and validated solution, and
• a pool of researchers to design and run field studies that test the effectiveness of our program in India and Kenya

Our primary technical objective is to be able to design reliable models of self-regulated learning (SRL) and metacognitive performance of children studying in low resourced settings in order to inform teaching-learning strategies that improve reflective and critical thinking in students.

Over the last 12 months, we’ve built our core capabilities i.e. to design, field-test and digitally distribute & grade contextual, valid and reliable assessments of metacognition and self-regulated learning (SRL), and have distilled existing SRL intervention literature into byte-size, student-friendly and actionable narratives that are personalized based on the learner performance, with an emphasis on helping students move from being tacit learners to reflective learners. 

We achieved this by solving two important problems related to modern assessment design:

1. Multimodal assessments that are platform agnostic, low-cost and can handle student responses via a variety of mediums, without burdening teachers with assessment design and grading
2. AI-enabled decision models that correctly account for time sensitive and sequential nature of self-regulated learning  

We believe that multimodal learning analytics are the future of assessments and data-driven student support, and can help bring a paradigmatic shift from the prevalent linear, navigated, and mastery-based approach to a non-linear, cyclic and life-skills based view of student outcomes.

Over the next one year, our goal is to continue scaling our existing self-regulation and metacognition assessments offering to 500K students, while developing new approaches to formative assessments and teacher support that address other key aspects of self-regulation such as affect (emotion regulation), cognition, and motivation.

With an open, interoperable and platform-agnostic assessment infrastructure, we aim to serve the socio-emotional learning measurement and improvement needs of 10 million students by 2028. Majority of this growth will come through government partnerships esp. in India where the demand for socio-emotional development is validated in a number of national  education missions, and the launch of a national education technology stack - NDEAR (National Digital Education Architecture) - has paved the way for innovative edtech solutions to rapidly scale up to 320Mn+ strong student base of the country. 

In Kenya, we will continue to expand our ongoing non-profit partnerships, further contextualising our solutions and building local teams to reach 1 million Kenyan students by 2028. 

Our progress measurement approach is a balance between user validation and short and long-term effects of Radics on student socio-emotional competencies.

Through structured field studies, and real-time data analytics we've established the following indicators of success:

• Validated demand for SEL assessments: 92% of teachers using Radics confirmed that our assessments will help them embed SEL instruction better in their daily lessons
• Feasibility of SEL assessments in low-resourced contexts: 96% of teachers agreed that multimodal assessments of SEL are feasible in their contexts
• Cultural, linguistic and age appropriateness: 80% teachers, 85% students and 80% experts in our ongoing engagements confirmed the context-fit of our assessment solutions
• Validity and reliability of assessment tools: using over 400,000 observations our metacognition and self-regulated learning assessments are found to be valid and reliable as per most recognized statistical tests

In the long-run, we plan to design longitudinal and experimental evaluation of our solution, prioritising following matrices:

• Shifts in student self-perceived capabilities to regulate cognition, emotion and motivation (from tacit to reflective learners)
• Shifts in student real-world outcomes such as math tests scores and language proficiency as a result of improved socio-emotional learning support

Self-regulation is proven to be a teachable skill and its specific aspects esp metacognition are shown to be strong predictors of reflective and critical thinking skills [CASEL; Zimmerman; Pintrich]. By periodically measuring a student’s self-regulation states and the extent of metacognitive activity when engaging in daily learning activities we can identify specific areas of improvement that can be addressed by applying established self-regulated learning strategies. Over long-term, students with better self-regulation and metacognitive abilities are likely to be reflective, life-long learners who can inclusively and equitably drive both personal as well as societal progress.

Radics is a collection of holistic learning measurement & improvement tools & services delivered through a rules-based conversational interface accessible via chat (WhatsApp, Telegram etc), and web. Additionally, teachers have a choice to conduct assessments via paper and auto-grade them using our custom computer vision enabled grading algorithm. 

Our core technology and methodology is applied to solve four important challenges related to modern assessments as follows:

1. Multimodal student responses: we’ve successfully implemented a set of natural language-driven assessment techniques that can engage students on a variety of question types such as multiple choice, fixed response, subjective response and audio responses. We transcribe audio responses to text in real-time and auto-grade responses for immediate feedback.

2. Computer vision for paper assessments: the traditional optical mark-up recognition based assessments (bubble sheets) lead to occasional drops in student attention. We’ve solved this problem by implementing a customized computer vision algorithm that leverages Python CV2 library to auto-grade natural student responses on paper (ticks, numbers, words, sentences etc)

3. Dynamic and sequential modeling of self-regulated learning processes: self-regulation is a complex process that involves four, inter-linked sub-processes - cognition, affect, monitoring and motivation. We leverage modern machine learning algorithms (dynamic bayesian networks) and AI-techniques (recurrent neural networks) to build reliable and accurate models of student self-regulation that can pinpoint strengths and areas of improvement in a timely manner

4. Narrative-based student and teacher advice: we deliver personalized student and teacher advice distilled from well-established strategies and the scientific literature that guide the development of socio-emotional competencies. We’ve designed a unique, hybrid approach to learning analytics that responsibly integrates a human expert-led advice with student-performance narratives generated with the help of ChatGPT. We subject all chatgpt-generated narratives to a five-point randomized evaluation (content, context, language, safety and trust) to ensure they are fit for student and teacher consumption.

Our approach to diversity, equity and inclusion is codified in our core values and lived through a set of early, but evolving, systems and processes.

DEI in our core values: our three core values are - Innovation with Responsibility, Excellence with Inclusion, and Ambition with Humility. As an edtech startup we believe technology can play a key role in not just improving learning outcomes but overall student-well being, and as we pursue our ambitions through innovation and excellence, doing so responsibly, with humility and inclusion is as important for us.

DEI in our practices: our commitment to responsible innovation is depicted in our data privacy and protection systems where we spend significant amount of our efforts on ensuring that our users' datasets are secure from malicious attacks. Additionally, we maintain a critically constructive approach to integrate generative AI tools such as ChatGPT. We practice excellence with inclusion by picking design challenges that solve for teaching-learning issues unique to low-resourced settings such as digital divide and relevance for paper-based learning tools. Lastly, while we aim to reach 10 million students by 2028, we acknowledge the fact that edtech is an enabler and is best applied to enhance the quality of human interactions.

DEI in our team: with technology at the core of our solution, we ensure equal or near equal representation from women in our technology and data science teams, and currently, 40% of our team members are female.

We are a for-profit social enterprise, and have been sustaining through B2B (mid-large education non-profits) and B2G engagements by offering learning measurement and improvement infrastructure tools and services. We’ve also served consulting assignments with the World Bank for building EdMIS capacity for regional governments in India. 

Both these revenue streams have enabled us to keep our solution free for students, teachers and schools who are currently underserved when it comes to data-driven socio-emotional learning support by government or NGO programs.

Digital and AI-enabled assessments of life skills is a growing solutions category within the $190 Bn+ global edtech market. We've validated this demand with mid-large non-profits and school districts with-in our target contexts, and have onboarded three institutions as strategic client partners. Since SEL research that advocates for multimodal learning analytics is in nascent stages, we aim to build on our early adoption experience and position ourselves as one of the top five SEL learning measurement and improvement solution providers in India and Kenya over the next 2-3  years.

We will fund this growth primarily via two avenues:

• Earned SaaS revenue: monthly recurring revenue of $0.10 per student per month from our B2B and B2G offerings
• Research grants: self-regulated learning is one of the most sought after research topics as it brings key education constructs (cognition, affect, motivation, metacognition) under a single umbrella and can greatly benefit from tools that reduce barriers to de-identified multimodal learning datasets. We've received product development and research grants in the past from Jacobs's Foundation (via MIT Solve Partner Challenges), and are one of top 64 solutions shortlisted for a $200K grant prize in the Learning Engineering Tools competition. This validates our relevance with-in the grant-maker community as a strong candidate for sustained research funding

We've sustained our operations through a mix of research grants and earned SaaS revenue. Below is a list of our major sources of funding so far:

• Research grant, $50K - Jacobs Foundation
• Earned revenue (consulting), $48K - World Bank
• Earned revenue (Saas), $55K - multiple non-profits clients (Nalanda Project, Learning Links Foundation - India)