Uniport Student, 18, Develops App That Makes Websites AI-Enabled

Trying this search enquiry on ChatGPT to book a flight for you, it will reply that it is unable to accommodate your request. However, it might tell you how to approach it, which airlines offer the greatest deals, and how to make a successful ticket purchase.

However, in order to enable ChatGPT to book flights directly, Obinna Chimdi, a student studying mathematics and computer science at the University of Port Harcourt (UNIPORT), Nigeria, has developed an AI agent named ChatATP. By serving as a bridge between ChatGPT and the airline’s website, ChatATP allows the chatbot to execute activities on the internet, allowing customers to complete tasks by merely typing commands.

Chimdi, who was only 18yrs at the time, struggled to find collaborators before creating ChatATP. “Everyone I wanted to collaborate with said no. “They were really busy,” he said over the phone to Techpoint Africa.

He pitched a simple idea to potential co-founders: create something that enables big language models like ChatGPT to do more than just communicate with users; it enables them to do activities that are not possible on chatgpt.com.

Developing ChatATP, Chimdi at 16 started studying programming. Stating that, he began programming with his dad’s cell phone. He taught himself as he went along, finishing his first assignment in three months.

He stated that his adoration for Mark Zuckerberg served as his driving force. According to Chimdi, “he was my idol.” He had to learn how because he had wanted to start his own AI company and he loved automations too. Motivated with this, he created Wall Street, a social media network aimed at bringing together businesspeople.

Chimdi’s most ambitious project to date is ChatATP. Because he had to upskill along the way, it took him six months to construct the platform.

Since he couldn’t locate any frontend devs, he had to study by learning how to be a backend developer. Not even UI and UX designers were available to him.

However, Chimdi’s annoyance with massive language models exceeded the challenge of locating co-founders.

He further stated that he realised that while ChatGPT and similar tools are powerful, they aren’t designed to solve problems end-to-end. For instance, they stop midway through if you ask them to evaluate data and email the report to someone. The remainder must be completed by copying, pasting, or doing it yourself.

He experimented with mixing various AI models and building execution layers on top of them as a result of the gap. The end product is ChatATP, an AI that does more than just respond. Consider it as providing ChatGPT with the necessary support to accomplish tasks.

How ChatATP operates which is an AI-powered HTTP, Chimdi believed it was a good idea to give AI the ability to “connect to the real world,” as he put it, because it is intelligent enough to understand humans or at least forecast what we might be saying.

He accomplished this by developing toolkits like ChatGPT, Gemini, or Claude that act as a barrier between a website and an LLM.

A toolkit functions essentially as a translator. It is a tiny, specially designed connector that makes certain activities visible. For instance, sending an email from Gmail, making a contact in HubSpot, checking travel availability, or looking for a university schedule. Every toolkit specifies the safety regulations it must follow, the inputs it requires, and the outputs it returns.

But the toolkits are only a minor portion of the problem. The Agents2 protocol, which Chimdi had to develop, is the focal point.

“HTTP for AI agents,” as he puts it, is a straightforward, all-purpose protocol that specifies how models and toolkits communicate with one another. Agents2 standardises how AI agents call tools, ensuring interoperability and scalability, much like HTTP standardised how browsers retrieve web pages.

ChatATP is primarily a tool for developers. In contrast to Decide or YarnGPT, ChatATP necessitates a certain level of technical expertise.

A user must first install and enable domain-specific tool kits, setup an API key from any major LLM provider, and select the model they wish to use to power their agent in order to get started. Following that, users of the website only input natural language questions, and the system uses the selected model and toolkits to generate results.

Using the ChatATP, the number of websites integrating ChatATP determines its level of popularity. Chimdi has only been able to onboard three people thus far: two classmates and his neighbour. He wants to draw in more users, but doing so would need him to spend more money on computation, which he does not currently have.

He may attempt to raise money, but investors would not be impressed by three users who also happen to be his classmates and his neighbours. Furthermore, it is nearly impossible to keep ahead of the competition in a world where artificial intelligence is developing on a daily basis without resources.

There is a lot of competition. Developers can already spin up autonomous agents, connect together numerous tools, and carry out intricate workflows with open-source projects like AutoGPT and LangGraph, frequently without the need for significant infrastructure.

Multi-model workspaces are being constructed by more recent entries like Lumio AI, which allows users to choose between GPT-4, Claude, Gemini, or Mistral and compare results instantly.

ChatATP lacks the robust development communities and substantial financing that these products have. Under these circumstances, three users are essentially a proof-of-concept.

However, ChatATP’s Agents2 protocol might set it apart. It provides a versatility that most rivals do not highlight by enabling developers to plug in multiple LLMs and toolkits in one location.

However, Chimdi will have to demonstrate that the product can operate at scale and attract consumers outside of his personal circle in order to make that advantage a true benefit. If not, ChatATP runs the risk of being eclipsed by more established, well-funded competitors and reduced to just another experimental AI effort.