Inspiration

Prior to the era of agentic AI, editing large dataset files was a complex and time-consuming process — even with the aid of LLMs — due to their inability to process enormous files effectively. But with the rise of agentic AI and SDKs that make it easier to develop such tools, a system capable of editing massive amounts of data is now a reality.

What It Does

Batchy is an agentic AI that detects the file format, loads it into a generic representation, and then performs changes to each item in the array one by one using generated queries. This allows full attention to each entry while preserving structure and formatting. Once processing is complete, the file is output with the requested transformations — fully formatted and clean.

How We Built It

By thoroughly reading the Google ADK SDK documentation, we were able to overcome the initial challenges of translating our theoretical concept into a working prototype. We iterated over early versions until we achieved a functional MVP.

Challenges We Ran Into

The documentation, while robust, is not well connected. It took a significant amount of time to study the examples in order to understand the file structure, agent workflow, and the many caveats and tweaks required to build a functional agentic AI system.

Accomplishments That We're Proud Of

We successfully built a bleeding-edge, state-of-the-art agentic AI capable of transforming how large datasets can be processed and edited with AI — a truly revolutionary step forward.

What We Learned

We learned the foundations of agent-based AI, the architectural principles behind these systems, and the complex inner workings that allow multiple AI agents to coordinate and act intelligently across tasks.

What's Next for Batchy

This is just an MVP — a demonstration — but the potential is enormous. We plan to:

  • Enable direct file loading from disk for gigantic datasets, such as those used for LLM training.
  • Support additional formats, including plain text, chunked and processed intelligently by AI.
  • Introduce a reviewer/critic agent for higher confidence and validation in transformations.

Built With

Share this project:

Updates