Tech Explained: Here’s a simplified explanation of the latest technology update around Tech Explained: The tech duo building an anti-ChatGPT for classrooms in Simple Termsand what it means for users..
Fermi, the startup founded by tech duo, is positioning itself not just as a place for students to use AI for learning and test preparation, but also as a tool for teachers to track where a student’s reasoning breaks down, flag weak areas, and step in earlier. For now, the startup’s focus is the Joint Entrance Examination (JEE) that selects students for top engineering schools in India.
Ranjan’s pitch to schools and teachers is straightforward: as GenAI tools spread in classrooms, students may appear to be doing better, submitting complete homework, but may actually be learning less.
He argues the problem is the chatbot interface that dominates AI, where a student can paste a question and get an answer instantly, removing the struggle that helps build understanding.
“Every kid is now turning in 100% completed homework with no errors…because the minute they run into any kind of struggle, there is ChatGPT to give the answer,” Ranjan said. “So, they can ask a question, but the AI within Fermi will never give you the answer. It will actually make you think.”
Unlike the subscription and course-led test-prep products in the market, including PhysicsWallah, Unacademy and Allen, Fermi is pitching itself as a tool for students inclined towards self-learning and practice-led study, rather than video-first learning and instructor-led classes.
The platform comes with a question bank focused on practising for Joint Entrance Examination (JEE), Ranjan said, and the company plans to expand to more courses over time, while staying anchored to competitive exam preparation. He also said building support for bilingual and local-language classrooms is on the roadmap, starting with English.
At the centre of Fermi’s pitch is the idea that students are relying heavily on general-purpose AI tools that hand out final answers, which can weaken the learning process.
For instance, an MIT Media Lab-led study in 20205 that tracked brain activity during essay-writing found the lowest engagement among participants using ChatGPT, and noted weaker recall and lower ownership of the work, compared with those using web search or no tools.
Evidence, data and learning outcomes
To support its claims, Fermi.ai has published a whitepaper based on student usage data from an August to December 2025 practice period focused on JEE aspirants. The dataset covers 79 students and 7,220 practice attempts on the platform, along with logs that capture how often students used hints and what kind of help they sought.
Fermi also tracked how well students understood specific concepts while they practised for the JEE examination. Since a single question can cover more than one concept, the platform ended up generating 15,047 concept-level scores from 7,220 practice attempts by 79 students.
To measure improvement, the paper assigns a single 0 to 10 score to each practice attempt, and then compares students’ initial scores with the later ones to see if they improved.
The paper reports that among the 57 students who did at least 10 attempts, the average score rose from 4.91 to 7.52, an improvement of 2.60 points. Among the 40 students who did at least 100 attempts, the average score rose from 4.97 to 7.79, an improvement of 2.82 points.
Ranjan said in an interview that in a Bengaluru pilot, students who began at about 2 on a 10-point scale improved by roughly 4.7 points by the end, while average scores rose about 2.6 points, and the company plans to validate these results later by comparing them with a control group once JEE outcomes are available.
Product design
Fermi.ai is also attempting to solve a practical constraint that often gets ignored in edtech demos: how students actually work through STEM problems. The platform is “stylus-first”, built around handwriting and a “smart canvas” so students can write equations and draw diagrams the way they would on paper.
Ranjan said the pilot that produced the 79-student results was run in Bengaluru, and the company is now running a pilot in Silicon Valley, starting another in North India, and is in discussions for Dubai.
“The go-to-market model and pricing are still being worked out. But for now, schools are being offered a free trial during the pilot phase for two to three months, including the question bank, co-created with our women and school’s educators,” he added.
The startup is currently funded internally by Meraki Labs, Mukesh Bansal’s venture studio. It is considering raising external capital in about six months, after completing more pilots and firming up its go-to-market strategy and pricing. Most of the tech team is based in Bengaluru, and the company is also starting to build a team in Silicon Valley.
Meraki Labs describes its typical first cheque as ranging from $250,000 to $5 million, alongside hands-on support in product, engineering, design and go-to-market. Apart from Fermi.ai, Meraki Labs’ portfolio companies include Nurix AI, Gigforce and NuShala.