Ariadne

Inspiration:

We saw a fundamental flaw in how everyone tries to solve inaccessibility. The crowd builds translation apps, screen readers, or PDF fillers. These are just digital bandages. They solve the word problem, not the comprehension problem.

Our inspiration came from a simple fact: bureaucracy is the greatest barrier to equality.

The knowledge and services for a better life—healthcare, legal aid, government benefits—are technically "available," but they are locked behind a "complexity barrier" of forms, jargon, and confusing websites. This system disproportionately fails 1.5 billion people with low literacy, visual impairments, or language barriers.

We didn't want to build another app. We wanted to build a master key.

What it does:

Ariadne is an AI-powered civic navigator that transforms any complex form, website, or document into a simple, guided conversation.

It's not a "form-filler." It is a personal guide through the labyrinth of bureaucracy.

SCAN: The user points their phone camera at any document (paper or digital) or website.
DECONSTRUCT: Our multimodal AI instantly identifies the document and understands the intent and logic of every field, not just the words.
TRANSFORM: It rebuilds the entire form into a simple, step-by-step conversation, delivered by voice and large, clear text.
GUIDE: Ariadne doesn't just ask; it explains. It turns complex jargon like "attestation of domicile" or "beneficiary" into simple questions like, "Who should get your benefits if something happens to you?"
COMPLETE: As the user answers, Ariadne populates the actual document in the background, handling all formatting and logic. It turns the specialist skill of "filling out a form" into the universal skill of "having a conversation."

How we built it:

For this ideathon, we designed the complete architecture. The "crowd" just hooks up an API. We designed a system for resilience and true understanding.

Front-End: A 100% accessibility-first mobile app (React Native/Flutter). The entire UI is navigable by voice, with high-contrast, large-print text as the default. This is non-negotiable. AI Core (The Brain): This is a multimodal model (like Gemini Pro Vision or GPT-4o). Its job is not just OCR; it performs Visual Document Understanding (VDU). It sees a form as a structured entity, identifying fields, checkboxes, and the logical relationships between them. The "Translator" Service (The Secret Sauce): This is a separate, fine-tuned LLM. Its only job is to take a "system" prompt (e.g., field: "beneficiary", type: "legal_person") and translate it into simple, Grade 5-level English, Spanish, Hindi, etc. This separation makes it robust. Back-End: A simple Node.js or Python service to manage the AI calls, maintain the "conversation state," and securely populate the final PDF.

Challenges we ran into:

The primary challenge wasn't the code; it was the logic.

Deconstructing "Messy" Documents: A clean, templated form is easy. A real form—scanned crooked, with coffee stains, or with a complex multi-column layout—is chaos. Teaching the AI to reliably parse any layout, not just perfect ones, was the central problem.
The "Jargon-to-Simple" Engine: This is harder than translation. Translating "liable" from English to Spanish is easy. Explaining what "liable" means to someone with no legal background, in a way that is accurate and simple, is incredibly difficult. This required iterative prompt engineering and fine-tuning.
Maintaining State & Logic: A form isn't a single question. It's a long, branching conversation. If a user ticks "Yes" on question 5, it might unlock questions 10-12. Building this logical state machine, managed by the AI, was a significant architectural challenge.

Accomplishments that we're proud of:

We are proud of identifying the real problem. The crowd was building faster horses; we designed a car.

Our main accomplishment is the Ariadne concept itself: moving beyond translation and OCR to build a true comprehension engine. We're proud of designing a system that delivers autonomy and dignity, not just "help."

The biggest "win" was the user flow. The moment we mapped out the "Scan -> Deconstruct -> Converse" loop, we knew we had a solution that was 10x simpler and more powerful than any other idea.

What we learned:

The most important problems aren't always the most obvious ones. The "crowd" is obsessed with headline-grabbing AI like image generation or chatbots for the privileged.

We learned that the most powerful and revolutionary use of AI is translation of complexity into simplicity.

The "digital divide" isn't about access to phones; it's about access to comprehension. We learned that "dignified design" is the future. Building for the most excluded groups (the non-literate, the visually impaired) creates a system that is fundamentally better for everyone.

What's next for Ariadne:

Ariadne is not just an app; it's a new accessibility layer for society.

Beyond Forms: We will expand Ariadne to deconstruct entire websites. Imagine turning a confusing 30-page government benefits portal or a dense healthcare insurance site into a 5-minute conversation.
Proactive Guidance: The app won't just be reactive. It will be a proactive agent. ("I see from this paper that you're eligible for this tax credit you didn't know about. Would you like me to help you apply?")
Platform & API: We will partner with governments, banks, and healthcare providers to offer Ariadne as a "bureaucracy-as-a-service" API. They can plug it in to instantly make their complex systems 100% accessible, without rebuilding them from scratch.