Ivy – 将 LLM 带给埃塞俄比亚的 3500 万线下学生
1 分•作者: zeshama•6 个月前
我是一名居住在亚的斯亚贝巴的开发者。当人工智能浪潮在云端蓬勃发展时，我所在地区的3500万学生却被远远抛在后面，因为他们无法获得可靠的互联网，也无力承担访问GPT-5 API的数据费用。<p>我创建了Ivy来解决这个问题。它是一个教育辅助工具，旨在完全在设备端运行（边缘推理）。<p>工作原理如下：<p>离线推理：在150美元的入门级安卓硬件上运行优化的本地LLM。<p>原生支持：集成对本地语言（阿姆哈拉语）的支持，以弥合文化差异。<p>自主性：即使电网或网关中断，系统也能正常运行。<p>我在这里记录了该系统的架构，以及一个在真实环境中运行的4分钟演示：https://builder.aws.com/content/39w2EpJsgvWLg1yI3DNXfdX24tt/aideas-ivy-the-worlds-first-offline-capable-proactive-ai-tutoring-agent<p>我目前正在参加一项全球AWS挑战赛的四分之一决赛，以获得扩大规模所需的资源。我正在寻求关于边缘推理逻辑以及这些设备连接2G信号时的同步协议的技术反馈。<p>很乐意回答关于在零连接环境下构建LLM基础设施的现实问题。
查看原文
I'm a developer based in Addis Ababa. While the AI Boom is happening in the cloud, 35 million students in my region are being left behind because they have zero reliable internet and can't afford the data costs of hitting a GPT-5 API.<p>I built Ivy to solve this. It’s an educational co-pilot designed to run entirely on-device (Edge-Inference).<p>How it works:<p>Offline Inference: Optimized local LLMs running on $150 entry-level Android hardware.<p>Native Support: Integrated logic for local languages (Amharic) to bridge the cultural gap.<p>Sovereignty: The system is designed to function when the grid or the gateway is down.<p>I’ve documented the architecture and a 4-minute demo of the system running in a real-world environment here: https://builder.aws.com/content/39w2EpJsgvWLg1yI3DNXfdX24tt/aideas-ivy-the-worlds-first-offline-capable-proactive-ai-tutoring-agent<p>I’m currently in the quarter-finals of a global AWS challenge to get this the resources it needs to scale. I’m looking for technical feedback on the edge-inference logic and the sync protocols for when these devices do hit a 2G signal.<p>Happy to answer questions about the reality of building LLM-infrastructure in a zero-connectivity environment.