Inspiration
Our first project idea did not go well, and for a while our whole team felt stuck. We were honestly close to giving up. Instead, we stepped back and decided to build something we actually wanted for ourselves: an AI that could do more than chat. We wanted something that could generate content, choose actions, and write files on its own. That need became Vyoma AI.
What it does
Vyoma AI is an AI task agent that can talk with the user, keep context during a session, and decide how to respond. It can either have a normal conversation, use an existing skill, or create a new skill from available tools and run it. It also works with assignment data, cleans and organizes information, and saves results into files like TXT, JSON, and CSV. The most interesting thing is that this has infinite scalability, the more tools you make the more control it has.
How we built it
Vyoma AI is an AI task agent that can talk with the user, keep context during a session, and decide how to respond. It can either have a normal conversation, use an existing skill, or create a new skill from available tools and run it. It also works with assignment data, cleans and organizes information, and saves results into files like TXT, JSON, and CSV.
Challenges we ran into
We built Vyoma AI using the Gemini SDK as the reasoning layer. Around that, we created a tool-based system in Python where the AI can call different utilities for file handling, data cleanup, formatting, and exporting results. We also built a skill runner that reads task steps, executes them in order, and passes data between steps.
Accomplishments that we're proud of
Our biggest challenge was that our original idea failed, which hurt our momentum early on. We also struggled with making the AI reliable, because it had to do more than generate text — it needed to choose the correct tools and use the right arguments. Another challenge was balancing flexibility and structure, so the system could create new skills without breaking existing workflows.
What we learned
We learned a lot about the Gemini SDK, tool-based AI workflows, higher-order methods, and designing reusable systems. Just as important, we learned a personal lesson: failure at the start does not mean the project is over. It can lead to a better idea if you keep going.
What's next for Vyoma AI
We learned a lot about the Gemini SDK, tool-based AI workflows, higher-order methods, and designing reusable systems. Just as important, we learned a personal lesson: failure at the start does not mean the project is over. It can lead to a better idea if you keep going.
Log in or sign up for Devpost to join the conversation.