Transmute

Transmute is an AI agent that turns forgotten "dark data" into a living knowledge map. It exposes contradictions and outdated policies, enabling organizations to act smarter and more sustainably.

Comment

Inspiration

We were initially drawn to the Cox Sustainability Track, but we wanted to weave in the event's magical theme. We chose to redefine "magic" as the process of taking something forgotten or decaying and transmuting it into something vital and new. In the tech world, this led us to dark data—the mountains of unused, duplicate, or forgotten files that clog servers and inflate the digital carbon footprint. Our goal was to take this "rot" and turn it into a living, manageable knowledge map that fuels sustainability.

Challenges: The Pivot

About 24 hours into the hackathon, we hit a wall. Our original concept relied on a multi-step AI process where the second stage proved technically unfeasible within our timeframe. We had to make the difficult call to pivot entirely. We salvaged our early momentum by refocusing our AI integration on the "Data Alchemy" concept, shifting our development from a broad mobile concept to a refined web-based intelligence tool.

How We Built It

Transmute functions as a digital laboratory that processes raw, "dark" data into a structured Wiki and Knowledge Map.

The Intake: Users upload archives (specifically ZIP files containing digestible text data). We built a system to handle POST requests that validate and process these files.

The AI Agent: We utilized Gemini to parse the content. The agent performs a triple-transformation:

Metadata Extraction: Generates a JSON representation of file types, names, and embeddings.

Visual Mapping: Creates a JSON-based graphical format to help users visualize how disparate data points relate.

Wiki Synthesis: Funnels the holistic data into a formatted Wiki page, providing a summary and actionable insights.

What We Learned

At the very start, we knew that our success would be determined by how well we can adapt and learn. Though we didn't expect this. Initially, we dove deep into React Native to build a mobile proof-of-concept, but we quickly had to pivot back to web-based components to better serve our "Data Alchemist" dashboard.

On the technical side, we mastered the art of prompt engineering. We learned how to query Gemini sparingly and strategically to ensure reliable data extraction—especially when moving from image recognition to complex text parsing. We also gained significant experience in managing file streams and structured JSON outputs to build a seamless bridge between "dark data" and clear, visualized knowledge.

On the visual side, we embraced the "Data Alchemist" aesthetic by crafting an immersive interface that bridges the gap between technical utility and digital magic. We developed a sophisticated dark-mode color scheme, utilizing deep charcoals to represent "dark data" contrasted against vibrant, glowing highlights to symbolize the transmutation process. By heavily leveraging CSS, we implemented complex box-shadows and glassmorphism to create a sense of depth and layered mystery.

Share this project:

Updates