Inspiration
We were inspired by the "Administrative Graveyard" in Indonesia. Currently, 45% of Family Hope Program (PKH) beneficiaries are mistargeted, and only 200k out of 1 million low-income applicants for national scholarships (KIP) could be funded in 2023. We realized that friction is the filter: people don't miss out because they are unqualified, but because the UI/UX of government bureaucracy is broken. High cognitive load forces users to navigate massive, fragmented systems under immense financial stress. We realized we don't need to change the users; we need to fix the interface.
What it does
Kertas transforms the system from a passive filter into an active bridge. It allows users to drop complex policy PDFs (like the LPDP Master's Application) directly into the app. The AI ingests the document, instantly creates a customized list of requirements, and provides a split-screen interface where the AI actively guides the application drafting process. Kertas speaks the language of bureaucracy so the user doesn't have to.
How we built it
We built Kertas using a secure "Zero-Leak Zone" architecture. Client Layer: A React UI that runs in the browser, providing a frictionless drag-and-drop interface.
Server Layer: A Flask/FastAPI Gateway that extracts text from the uploaded guidebook and handles context injection.
AI Engine: Mistral AI, which processes the prompts and returns JSON arrays containing chat responses and document drafts.
Secure Output: The final PDF generation happens purely on the client-side to ensure sensitive data never hits the cloud.
Challenges we ran into
Our biggest challenge was building trust. Citizens are highly hesitant to use AI because they don't want to upload highly sensitive personal data (like their NIK/ID) to cloud servers. We had to figure out how to provide the power of an LLM without compromising data security.
Accomplishments that we're proud of
We successfully built the "Zero-Leak Zone". The AI generates drafts using placeholders, and sensitive identifiers are typed locally by the user directly in their browser. The final PDF is generated client-side, achieving zero data leakage.
What we learned
We learned that administrative rejection rates are staggering—up to 45% of applicants are rejected simply because their documents miss hidden formatting rules or specific grading criteria buried in guidebooks, not because they lack merit.
What's next for Project Kertas
Our IT Implementation Roadmap includes four main
- Action Planner: Creating a step-by-step roadmap to transform "what if" into "how to".
- Evidence Graph: Mapping exact statuses and citing specific pages from policy documents for 100% transparency.
- Frictionless Input: Implementing on-device OCR for zero-leak scanning and better document tracking.
- Sustainability: Establishing B2B licensing for institutions to scale their citizen reach globally.
Log in or sign up for Devpost to join the conversation.