How Fermi is reimagining AI for the modern classroom

When schools across the globe shut for the summer holidays, an unusual trend emerges in the digital world. Traffic on ChatGPT, one of the most popular artificial intelligence (AI) chatbots, begins to plummet.

This observation, noted by Peeyush Ranjan, Chief Executive Officer of edtech startup Fermi.ai, highlights an increasing dependence on generative AI tools by students. The concern with this trend is that students are using these tools to find immediate answers to difficult homework problems rather than engaging with the underlying material.

Ranjan, a former Google and Airbnb executive and former CTO of Flipkart, explains that, while the answer might arrive via a chatbot, the learning often does not—because true understanding requires a level of mental effort that shortcuts simply cannot provide.

Led by Ranjan, Fermi.ai is attempting to change this dynamic by shifting the focus from speed to reasoning. The global edtech startup has launched its AI-driven platform in India and US through subsidiaries.

The platform targets high-school STEM subjects beginning with mathematics, physics, and chemistry.

By providing stepwise coaching and deep analytical insights for both students and teachers, the startup aims to ensure that technology supports thinking rather than replacing it.

Fermi is headquartered in Singapore, and has offices in India (Bengaluru) and the United States. The initiative has emerged from the Meraki Labs ecosystem with Mukesh Bansal.

Meraki Labs is a venture studio focused on incubating and scaling next-generation technology startups. Ranjan joined Meraki as a partner last year.

Productive struggle

Unlike general AI tools designed to provide instant solutions, Fermi.ai acts as a digital tutor that prioritises what the company calls the “productive struggle”. 

Fermi’s platform is rooted in the belief that “mental struggle” is necessary for cognitive development. The current reliance on general-purpose AI is leading to a weakening of cognitive skills, as students bypass the brainwork required to solve complex problems.

“Learning requires you to go through that struggle and really make that change in your brain. The more people use AI, the more their cognitive skills actually weaken because AI gives you the answer which otherwise would have come out of your brain,” says Ranjan, explaining the core objective of the platform.

To combat this challenge, Fermi models the behaviour of an “exceptional teacher” who provides just enough support to keep a student moving forward without revealing the final solution.

The platform provides an adaptive real-time AI tutor which guides students to the next level of understanding only when they are genuinely stuck.

If a student expresses confusion, the AI does not provide the answer immediately but instead asks clarifying questions about the concepts involved, encouraging the student to re-evaluate the problem from another perspective. 

“It’s like a great teacher who’s sitting with you and that great teacher who’s sitting with you will never give you the answer. It will make you feel supported such that you come to the answer yourself,” Ranjan notes.

The platform also has a handwriting-first interface, which recognises that STEM reasoning is a tactile process. Students are encouraged to use a stylus and digital ink on a ‘smart canvas’ to draw free-body diagrams, write chemical equations, or solve integrals.

This approach aims to eliminate the friction and distraction of prompt engineering, a common barrier when using general chatbots, allowing students to focus on the problem at hand.

The platform is designed to be accessible across a range of hardware, from high-end iPads to affordable Android tablets and even Wacom drawing tablets plugged into laptops. Supporting these interactions is a deep concept graph and a curriculum-aligned question bank that covers major examinations such as JEE, AP, and IB. 

Data-driven insights 

Apart from the student experience, Fermi also provides data-driven insights to educators. 

When homework is completed through the platform, Fermi’s ‘Classroom Command’ feature gives teachers visibility into exactly where a student’s reasoning has faltered and identifies specific patterns of struggle that were previously impossible to track in a traditional setting.

“Fermi.ai isn’t here to give answers; it’s here to provide the map and the mirror, showing students how they think and giving teachers the visibility to lead them back to the path of mastery,” says Mukesh Bansal, Partner at Meraki Labs.

The approach is measurable, says the company. A three-month pilot involving 79 students and over 15,000 concept-tests indicated that students who initially struggled with scores of 2 out of 10 saw their averages improving to 6.7 out of 10.

Furthermore, students who completed 100 or more practice attempts showed a 21% drop in their reliance on hints, suggesting that repeated and guided practice fosters genuine independence, says the company. 

Tech architecture 

Fermi.ai does not rely on a single foundational model but uses various frontier models including GPT 5.2 and Gemini 2.5. It also uses a proprietary benchmarking tool called ‘GE bench’ to evaluate which models perform best for specific subjects or problem types. This allows the platform to remain agile in a fast-moving AI landscape, drawing in the best developments from across the industry to enhance the tutor’s performance, says Ranjan.

The tech team behind Fermi’s architecture includes veterans from companies such as Google, Walmart, Vedantu, and Cuemath, combining technological expertise with education domain knowledge.

Market strategy 

Fermi is currently targeting the two largest after-school edtech markets in the world: India, which is valued at $8 billion, and the United States, valued at $5 billion, according to Ranjan.

The company is running pilots in Silicon Valley, in the US, and various regions across India, including Bengaluru and parts of northern India.

While the platform is open to direct adoption by students, the company views schools as essential B2B partners for reaching a large number of learners.

Ranjan says schools have a strong incentive to align with this technology because they care about exam outcomes and the development of real mastery.

As the company moves towards a full commercial launch, it plans to introduce flexible pricing models, including subscription packages and top-up options for solving specific sets of questions.

While the current focus is on high-school STEM, the learning platform is designed to scale quickly to other scenarios. Ranjan notes that it took only “two weeks” to add chemistry to the platform.

Future plans include expanding into college-level courses and professional re-skilling.

Despite the presence of giants like Google and OpenAI in the AI-led education space, Fermi believes its specialised approach to friction-less, reasoning-based learning sets it apart.

“We will make learning hard things easier. You take quantum physics. You take deformable mechanics… We will make all of these things easy for you,” he says, adding, “The way we are solving is not necessarily that easy for an incumbent player to adopt without giving up their core."