Project name

Famora AI

Elevator pitch

An AI family assistant that turns bills, appointments, and everyday requests into trackable commitments and priorities, powered by NVIDIA Nemotron on Nebius Token Factory.

Project description

Inspiration

Family information is often scattered across bills, school notices, appointment letters, receipts, and conversations. Remembering every due date, tracking household expenses, and coordinating responsibilities can become difficult.

We built Famora AI to bring this information together and help families turn everyday information into clear, actionable commitments.

What it does

Famora AI is an AI-powered family assistant that helps households organize their information in one place.

Natural-language AI assistant: Ask questions about family commitments and use supported requests to create or update records.
Document intelligence: Upload bills, invoices, school notices, and appointment documents. The system extracts proposed information for review before creating a commitment.
Commitment tracking: Organize bills, appointments, tasks, renewals, and deadlines.
Expense management: Record expenses and view category-based totals and summaries.
Family timeline: View upcoming commitments and deadlines chronologically.
Dependencies and priorities: Track tasks that must happen before other events and identify commitments that need attention.

How we built it

The frontend uses React, TypeScript, and Vite. The backend uses Python and FastAPI, with PostgreSQL for persistent storage.

The AI agent uses NVIDIA Nemotron 3 Super 120B A12B through the Nebius Token Factory API for natural-language reasoning and tool-supported interactions.

Instead of relying on a chatbot response alone, the agent can request registered application tools. The backend validates the requested inputs, executes supported actions, stores the resulting records, and returns the actual results to the agent.

For document processing, the system extracts text or analyzes supported images and PDFs to propose structured fields. Users review the extracted information and confirm it before a commitment is created.

The application also uses deterministic backend logic to calculate priorities from dates, statuses, and dependencies.

Challenges we worked on

One challenge was connecting natural-language requests to reliable application actions. We needed to validate model outputs, handle missing or invalid information, and ensure that the agent used the correct family context.

Document extraction presented another challenge: the information found in a bill or appointment notice must be presented clearly and reviewed before it becomes a saved record.

We addressed these challenges with structured extraction, validated tool execution, database persistence, and explicit confirmation for document-based commitments.

What makes Famora AI different?

Famora AI goes beyond answering questions. It connects family context, document extraction, structured records, and application tools so users can move from everyday information to trackable actions.

Our goal is to make family coordination simpler, more organized, and easier to maintain.

What's next?

We plan to improve document extraction accuracy, make extracted fields easier to correct, strengthen contextual memory across conversations, and expand family coordination features while improving reliability and testing.