Fourteen-year-old Arjun Shah, an Indian-origin entrepreneur from San Jose, California, claims to have developed a solution to one of the biggest challenges facing modern artificial intelligence—high AI token costs. Through his startup, Supercompress, Shah says he has created a tool that reduces the number of AI tokens required by eliminating unnecessary context while preserving essential information.
Addressing rising AI costs
Every interaction with an AI model consumes tokens, and the amount increases as more context is provided. Since processing large amounts of context is expensive, businesses using AI at scale often face rising operational costs.
Explaining the problem to India Today Tech, Shah said AI applications frequently send far more context than necessary, much of which is redundant, forcing companies to pay for processing information that contributes little to the final response.
According to Shah, Supercompress analyses both the user's query and the accompanying context, identifies the most relevant information and removes the rest. He claims the tool reduces context by an average of 65 per cent while retaining more than 98 per cent of the critical content.
Describing the technology, Shah said Supercompress is a neural network framework structured similarly to the human brain, with interconnected nodes designed to identify important information efficiently.
He explained that the system reads the entire context, determines which portions are relevant to a specific query and sends only the essential information to the AI model.
In a demonstration, Shah claimed the tool compressed context by 97.5 per cent while producing the same output as the original input, but with faster processing and significantly lower token consumption.
At the time of writing, Supercompress reportedly has around 150 users and is also available as a plugin for AI coding agents.
From coding at seven to building AI
Shah said he began coding in Python at the age of seven after being introduced to programming at school, but largely taught himself through experimentation.
Interestingly, he says he stopped coding by the age of nine because he disliked memorising programming syntax, although his passion for creating new products remained.
The emergence of AI-assisted coding tools such as Claude Code and Cursor enabled him to continue building software without focusing heavily on writing code manually.
He said he began using Cursor around two years ago and created an AI-powered plant identification application for a school science fair.
Shah added that he then spent an entire summer studying neural networks before eventually developing one of his own.
Rather than learning programming languages in depth, he said he prefers studying the underlying concepts, including mathematics, algebra and the technologies that power neural networks.
Y Combinator application
Confident in his product, Shah has applied to Y Combinator, one of Silicon Valley's best-known startup accelerators.
If selected, he would receive USD 500,000 (approximately Rs 4.8 crore) in funding along with access to Y Combinator's network of founders and mentors.
Y Combinator has previously backed several successful startups, including Airbnb and Reddit. The accelerator has also accepted exceptionally young founders in the past, including teenagers approved during OpenAI CEO Sam Altman's tenure as YC president.
Shah said his parents were originally from Gujarat before emigrating to the United States, where they later became US citizens.
He also credited his mother for supporting his journey in technology and noted that she co-founded his previous venture, therooted.ai, a retrieval-based platform that draws on the Charaka Samhita and other traditional texts to suggest remedies for common ailments.
