Inspiration
Keeping track of valuable items is often a tedious, manual process. Holos was inspired by the need to simplify home inventory and asset management, replacing spreadsheets and manual data entry with a streamlined, visual approach.
What it does
Holos is an AI-powered room scanner and intelligent home cataloging application. By simply snapping a picture of a room or a specific object, the system automatically handles:
Item Identification: Extracts the name, category, make, and model.
Automated Valuation: Provides real-time market value estimates based on the item's condition.
Smart Parsing: Uses context-aware extraction to identify specific details like book titles, authors, ISBNs, and physical dimensions.
How we built it
The project utilizes a modern, cloud-native tech stack designed for speed and deterministic results:
AI Engine: Powered by Gemini 2.5/1.5 Flash Vision models, utilizing custom temperature constraints to ensure structured JSON data extraction.
Backend: Built with Flask (Python 3.10+).
Frontend: Developed using vanilla JavaScript and custom CSS, featuring an SEO-optimized landing page.
Database & Auth: Integrated with Supabase (PostgreSQL) for secure user profiles and image tracking.
Challenges we ran into
While not explicitly detailed in the "Setup" or "Highlights," the technical configuration suggests hurdles in:
Deterministic Extraction: Ensuring the AI consistently returns valid JSON for structured data.
Environment Configuration: Managing complex integrations between the Gemini API, Supabase, and the Flask environment.
Staging & Version Control: As noted in the project history, resolving technical issues with Git staging and repository management.
Accomplishments that we're proud of
Seamless Integration: Successfully bridging high-level Vision models with a functional web backend.
Automated Workflow: Moving from a raw image to a fully parsed, valued, and categorized database entry without manual user input.
SEO & Design: Creating a "beautiful" custom CSS interface that is also optimized for organic search traffic.
What we learned
The development process highlighted the power of Vision-Language Models (VLMs) in replacing manual data entry. The team gained experience in:
Configuring cloud-native storage and authentication via Supabase.
Implementing temperature constraints in LLMs to maintain data integrity.
Structuring a Python-based microservice architecture for AI applications.
What's next for Holos
The project is moving toward a broader consumer release:
Mobile Launch: Coming soon to the iOS App Store and Google Play.
Scaling: Further refining the "Holos | AI Room Scanner" to handle more complex domestic and commercial environments.
Log in or sign up for Devpost to join the conversation.