Inspiration
We started Education Harness Neko with a simple question: what if AI could do more than just answer students' questions?
Most AI education tools feel like chatbots with a different interface. They can explain a concept, but they rarely understand the learning process or know what to do next.
We wanted to explore a different approach: building an AI Agent that can understand learning goals, break down tasks, use tools, evaluate its own responses, and adapt to students over time.
As computer science students, we also wanted to build something we could actually use on our Android devices, not just another AI demo running in a browser.
What it does
Education Harness Neko is an AI-powered educational application built around an intelligent Agent system and a structured Agent Harness.
The application helps students ask questions, understand difficult concepts, organize learning tasks, and receive personalized guidance.
Instead of relying entirely on a single large language model, we designed a system where AI Agents can work with educational knowledge, tools, conversation history, and evaluation mechanisms.
The Harness manages how Agents operate, including context management, tool execution, response validation, and task workflows.
On the user side, we built a Flutter application for Android, making AI-assisted learning accessible through a familiar mobile experience.
Our goal is to make the AI feel less like a search box and more like a learning companion.
How we built it
We divided the project into three major parts: the mobile application, the AI Agent system, and the Harness infrastructure.
For the frontend, we used Flutter and Dart to develop a cross-platform mobile application, with Android as our primary target. We focused on responsive interfaces, conversation interactions, and integration with the Android ecosystem.
For the backend, we worked with Python and FastAPI to handle API requests, user interactions, and communication with AI models.
The most interesting part was the Agent Harness engineering.
We designed a structured execution workflow that connects language models with external tools, knowledge retrieval, context management, and evaluation components.
The Harness provides the environment in which Agents operate. It controls how tasks are processed, how tools are called, and how results are returned to the application.
We also explored Retrieval-Augmented Generation (RAG) to connect AI responses with educational knowledge rather than depending entirely on model memory.
Throughout development, we used AI-assisted coding tools to accelerate implementation, debug integration problems, and experiment with different Agent architectures.
Challenges we ran into
One of our biggest challenges was realizing that building an AI Agent is very different from simply connecting an application to a language model API.
An Agent might generate a reasonable response while still choosing the wrong tool, misunderstanding the context, or failing halfway through a task.
We spent considerable time thinking about how the Harness should manage execution states, tool permissions, conversation context, and error recovery.
Another challenge was integrating AI workflows into a mobile application.
Flutter made cross-platform development easier, but Android introduced practical issues involving application lifecycle management, network connectivity, asynchronous requests, and UI state synchronization.
We also found that evaluating AI responses is surprisingly difficult. A fluent answer is not necessarily a correct or useful answer, especially in an educational setting.
These problems pushed us to think more seriously about reliability, evaluation, and the engineering behind AI applications.
Accomplishments that we're proud of
We're proud that Education Harness Neko became more than a simple chatbot prototype.
We brought together Flutter mobile development, Android application engineering, AI Agents, backend services, and Harness architecture into one project.
The biggest achievement for us was moving from individual AI API calls toward a structured Agent execution system.
We also gained practical experience designing an AI application from both sides: what students interact with and what happens behind the scenes.
Seeing the mobile interface communicate with our backend and trigger Agent workflows was one of the most rewarding moments of the project.
It may not be a massive platform yet, but it represents a meaningful step from experimenting with AI models to engineering actual AI-powered software.
What we learned
This project changed how we think about building AI applications.
We learned that the language model is only one part of an Agent system. The surrounding engineering matters just as much.
A well-designed Harness can make Agent behavior easier to control, observe, evaluate, and improve.
We also learned a lot about Flutter, Dart, Android development, backend integration, asynchronous programming, and API design.
More importantly, we began to understand the difference between making an AI demo work once and building an application that people can use repeatedly.
For educational AI, accuracy, context, and user experience are all essential.
What's next for Education_harness_neko
Our next step is to strengthen the Agent Harness and expand the educational capabilities of the application.
We plan to introduce more specialized Agents for learning assistance, task planning, knowledge retrieval, and answer evaluation.
We also want to improve the Harness with better execution tracing, automated testing, model evaluation, and more reliable tool orchestration.
On Android, we plan to continue improving the Flutter interface, local data management, notifications, and integration with native platform capabilities.
In the longer term, we want to explore multimodal interaction, personalized learning memory, and more advanced Agent collaboration.
Education Harness Neko is still evolving, but the direction is clear: we want to build an AI education system where intelligent Agents are supported by solid engineering, not just clever prompts.
