Jinso Labs

Last week, our team attended the ARTESOL Annual Convention in Argentina, one of the region’s most important gatherings for teachers, and an event we were proud to sponsor. Between sessions, we crossed paths with Andrea Trujillo, an ELT teacher from ASICANA with 19 years of experience, who has been using Jinso with her students. We asked her how the experience has been going. Here’s what she had to say ↓

19 countries. 100,000+ learners. One consistent lesson: AI in education only produces real results when three conditions are met. → The first: the AI knows what each student needs to learn, not just what they're asking. That requires a competency layer connected to the institution's actual curriculum, not a generic content library. → The second: teachers lead the process. The institutions where outcomes improve are the ones where AI gives teachers better signals, not where it operates independently of them. → The third: the institution can measure what happened. Engagement data alone doesn't answer the question that matters. Assessment results, skill mastery, progress over time — that's the evidence that justifies continued investment. These three conditions aren't complicated. But they require a different kind of deployment than most institutions are doing. Most institutions are buying AI features. The institutions getting results are building AI infrastructure.

Most institutions deploying AI in classrooms are quietly reducing what students learn. They don't know it yet. The product looks identical from the outside. A Wharton study proved exactly this. ~1,000 students. Same AI model. Two different setups. Group A used an AI that answered questions directly. Group B used an AI prompted to guide, never to give the answer. While using the AI, both groups improved: → Group A: +48% on problem-solving → Group B: +127% Then they took the AI away and tested students alone. → Group A: scored 17% below students who had never used AI at all → Group B: the negative after-effect was largely mitigated The researchers called it a "crutch effect." Without guardrails, students stopped thinking and started copying. The tool looked like it was working the entire time. It was only when they took it away that the real damage became visible. This is exactly Jinso’s differentiator: structured AI, designed around learning outcomes — not just usage. Is your institution measuring what students actually retain, or just how much they interact with the tool?

Hosted last week at the U.S. Embassy’s new Mexico City compound, ABLA Express 2026 brought together State Department partners and BNC leaders from across the Americas around a central question: how do binational centers remain essential in a changing education landscape? The message across the program was clear: binational centers are more than language institutions. They are trusted bridges between U.S. diplomatic priorities, local community needs, private-sector capacity, and evolving workforce and education systems. As education adapts to new technologies, workforce demands, and expectations for measurable impact, that role is only becoming more important. We were grateful to support the gathering, contribute to the conversation with insights from our founder, Jay Lee-Gopalan, on building AI infrastructure for scalable, accessible, and locally relevant education systems, and host the closing Jinso Taco Night event. Thank you to the organizers: ABLA - Association of Binational Centers of Latin America, IMARC | Instituto Mexicano Americano de Relaciones Culturales, A. C., Instituto Mexicano Norteamericano de Relaciones Culturales, Luis Alfredo Medina Córdova, Claudia Zuniga and Ana Laura, the speakers and everyone who joined us in CDMX 🇲🇽 And a special thank you to Katherine Carabajal for helping us share the incredible impact Jinso has been making at ASICANA.

You have heard us call ourselves an AI learning infrastructure company. But what does that actually mean in practice? We built the layer that connects an institution's curriculum, student data, competency standards, and existing systems into something that can drive and measure learning outcomes. That is a different problem to solve than building a better LMS or a smarter AI tutor. In practice, this means engineers working inside institutions rather than handing over a product and a manual. Each deployment is shaped by the institution's: → Curriculum and competency standards → Existing systems and workflows → Way of defining and measuring progress What comes out of that process is a system that knows each student's current level, identifies the specific concept blocking their progress, personalizes the practice needed to move past it, and surfaces that data in real time to teachers and administrators. Across any subject, any competency framework, at institutional scale. Learning infrastructure is what you get when AI is built around how an institution actually works rather than asking the institution to work around the AI.

Students using AI in their studies scored 6.7 points lower on exams than students who did not use it at all, according to a study by the University of Bremen and Concordia University. That finding is one of the most uncomfortable data points in education research, because it runs directly against what institutions expected when they started adopting AI tools. The absence of structure is what turns a learning tool into a shortcut: → No competency framework → No guided practice layer → No assessment loop Students are offloading the work of thinking to AI rather than building real understanding. A 2026 report from the University of Technology Sydney made the consequence more specific: unstructured AI use is widening equity gaps. Students with strong prior knowledge can leverage AI effectively. Students without that foundation — the ones who need the most institutional support — are the ones falling furthest behind. Institutions bought AI adoption without buying the structure that makes AI work. Real results come from AI where the pedagogy and instructional architecture are built in.

Most students who struggle aren't missing effort. They're missing one concept nobody realized they never learned. Every subject has a hidden architecture. Skills that only make sense once another skill is already in place. You can't read a graph without understanding scale. You can't write an argument without knowing what a claim is. You can't solve for an unknown without understanding what an equation is actually saying. When a student hits a wall, it almost never happened at the wall. It happened three steps earlier, at the concept that should have been the foundation for everything that followed. Research shows students consistently overestimate their own understanding and can't identify the prerequisite knowledge they're missing. The gap is invisible to the person experiencing it. Which means asking students what they don't know produces the wrong answer almost every time. This is why the order in which things are learned matters as much as what is learned. A competency graph maps that order. Every skill connected to the skills it depends on and the skills it unlocks. Not a flat list of topics, but a structured progression where mastery at one level becomes the foundation for the next. It makes the hidden architecture of any subject visible, measurable, and actionable. When you know the architecture, you can find the exact block. Not the subject a student is struggling with. The specific concept that, once understood, unlocks everything stacked above it. That's what personalized learning actually means. Not a different interface. A different starting point.

The average school district spends $2.3 million annually on technology tools that deliver minimal return on investment. (SETDA) That figure understates the real cost. Teachers lose an average of 47 minutes per week to technology tasks that have nothing to do with instruction. IT departments spend 34 percent of their time just keeping fragmented systems running, roughly $180,000 in labor annually, per district, before a single student opens a platform. Fragmentation is the cost. The more platforms an institution adds, the more overhead it creates and the less visible the outcomes become. Every disconnected platform creates its own integration overhead, its own training requirement, its own data silo. Administrators cannot see who is falling behind. Leadership cannot justify the spend at budget review because the data to prove value does not exist. Jinso resolves all of it at once. A fully managed LMS, deep integrations across existing systems, unlimited teacher training, and independent outcome data. One contract, replacing seven vendor relationships, and producing the data institutions need to justify every dollar spent.

322,000 AI conversations. 2,898 students. One national pilot. Here are 4 things we learned and what the data actually showed: 1) Students don't need another app. 85% activated Jinso within 48 hours — because it runs entirely on WhatsApp. No downloads, no friction, no IT onboarding. The tool met students where they already were. 2) Consistency of use is what drives results. Classrooms that fully integrated Jinso averaged 87.3 on their English exams. Those with limited adoption averaged 81.5. That +5.8-point gap didn’t come from the technology alone; it came from teachers making it a habit, not an option. 3) Scale doesn't have to mean losing quality. This ran across 13 institutions and 56 classrooms simultaneously, over 4.8 months of continuous usage. The learning was personalized. The results held across the board. 4) The data changed the conversation. MESCYT's own Quantitative Impact Evaluation (2025) concluded that the results support transitioning Jinso from an optional resource to a fully integrated institutional requirement. In the ministry’s own words. If outcomes matter to you, this case study is worth a read. 👉 Check the full MESCYT Quantitative Impact Evaluation study here: https://buff.ly/9SwLQNV

When students are placed in the wrong course, the damage is immediate: they fall behind, disengage, or drop out entirely. Placement decisions are still periodic, which means errors compound before anyone catches them. The default response is to expand supply through more programs, more hiring, and higher cost. Institutions are spending more to solve a problem that better data could fix. That starts with continuous visibility into student performance and system-wide demand. Jinso's automated placement assessment does exactly this. Every student enters at the right level, every cohort is tracked in real time, and institutional decisions are driven by live data rather than end-of-semester reports.