Zipo AI

Overview

Southeast Asia faces a dual crisis in education: a significant Quality Gap and an Accessibility Gap. Out of a population of 685 million, approximately 47% live in rural areas with limited or unstable internet access. While students in major urban centers enjoy advanced AI-powered learning tools, those in remote provinces are often left with static textbooks that may not even be available in their native languages, creating substantial linguistic and infrastructure barriers.

The region’s vast diversity of over 1,000 regional languages demands a solution that goes beyond simple, literal translation. Without a digital intervention that is both culturally nuanced and adaptive to low-connectivity environments, this educational divide will continue to limit the ambitions and potential of millions of young talents across Southeast Asia.

The Solution:

Zipo is an AI-Native, Agentic Learning Platform designed as a Progressive Web App (PWA) featuring an AI tutor that Adapts, Explains, and Illustrates. Zipo fundamentally reimagines the learning process by transforming passive reading materials into active, conversational dialogues. It pairs verbal explanations with a real-time generative whiteboard, ensuring deep understanding even in offline environments through its innovative `.zipo` module system.

Zipo utilizes an AI-Native Agentic RAG architecture powered by the SEA-LION large language model (aisingapore/Gemma-SEA-LION-v4-27B-IT). We integrated SEA-LION not just as a text generator, but as a Visual Orchestrator. By exposing our frontend canvas library (react-konva) as a set of tools through Native Function Calling, the model autonomously “drives” the application, deciding how to layout information visually while it speaks

Key Features:

1. Generative Learning Canvas: Zipo doesn’t just respond with text; it dynamically draws diagrams, flowcharts, and illustrations on a digital canvas. These visuals are perfectly synchronized with the AI’s verbal explanation (TTS using Google Chirp-3), providing a multi-sensory learning experience.
   
2. Agentic RAG (Retrieval-Augmented Generation): Unlike standard RAG systems that passively receive context, Zipo acts as an agent. It autonomously decides when it needs more information from uploaded documents (PDFs/Text) to answer a specific user query, triggers the retrieval tool, and then synthesizes a comprehensive response.
   
3. PWA & .zipo Offline Modules: To solve the infrastructure gap, Zipo can package an entire AI generated lesson—including visual commands, synthesized audio, and transcripts—into a highly compressed, portable  .zipo  file. These modules can be downloaded and shared via WhatsApp or Bluetooth) to be played back with full interactivity entirely offline.

Conclusion

During our development and demonstration phase, Zipo has shown that AI can be a powerful tool for accessibility, not just a luxury for the connected. By utilizing the SEA-LION model, we achieved high accuracy in understanding hyper-local Indonesian contexts that Western-centric models often overlook. Our agentic RAG approach demonstrated that multi-turn reasoning significantly improves the relevance of AI explanations compared to traditional, passive RAG.

What’s Next for Zipo:

1. Expand Language Support: Our roadmap includes covering more of the 1,000+ languages in the Southeast Asian region to ensure no learner is left behind.
2. Collaborative Learning: We plan to implement real-time synchronization features, allowing multiple students to interact and learn together on the same generative canvas.
3. Teacher Authoring Tools: We aim to empower educators by providing intuitive tools to create and distribute complex, customized .zipo modules easily.

Zipo is more than just an application; it is a scalable and sustainable commitment to bridging the educational gap in Southeast Asia. We are moving from “Access to Information” to “Access to Understanding,” making world-class personalized learning a reality for everyone, everywhere.

About the Team

Behind Zipo, we are team AI-Native Agentic RAG Vertical B2B SaaS, composed of 5th-semester students from the School of Electrical Engineering and Informatics (STEI-K) at Bandung Institute of Technology (ITB).

Team Members:

1. Khalfani Shaquille (Project Lead): Information Systems and Technology at ITB. Focused on AI agents, orchestrations, and backend architecture. (khalshaquille03@gmail.com)
2. Varel Tiara (Collaborator): Informatics at ITB. Specialized in Frontend Web Development. (vareltiara@gmail.com)
3. M Ikhbar A (Collaborator): Information Systems and Technology at ITB. Expert in Data Science and Product Management. (mikhbar17@gmail.com)

Our team is driven by a shared mission to provide world-class, personalized education delivered in native languages. We believe that technology should be an equalizer, not a divider. By participating in the PAN-SEA AI Developer Challenge, we committed ourselves to ensuring that even the most remote learners have access to the same educational quality as those in urban centers.

We envision Zipo as the standard for adaptive learning in the region—a platform that respects cultural diversity and thrives in environments where connectivity is a challenge. Our goal is to empower the next generation of Southeast Asian talent through advanced, accessible AI-driven generative learning.

Appendix

Try Zipo: zipoai.xyz

Product Demo Video: https://youtu.be/Tj8i6i1StDg