Inspiration In the current AI landscape, most assistants are "talkers," not "doers." We were inspired by the potential of Google Cloud’s Agent Builder to create something that bridges the gap between conversation and commerce. We wanted to build SwiftOperate to prove that an AI agent can handle high-stakes operational tasks—like inventory management or emergency dispatch—with the speed and precision of a human expert.
What it does
SwiftOperate is an autonomous operational agent that listens to user intent and executes complex workflows across multiple platforms. Instead of just answering questions about data, it interacts with APIs to:
Update real-time databases.
Trigger automated cloud functions.
Coordinate between different team communication channels (like Slack or Email) to ensure an action is completed from start to finish.
How we built it
We built the core of SwiftOperate using Vertex AI Agent Builder, utilizing Gemini 1.5 Flash for its low latency and high performance.
Playbooks: We designed modular playbooks to handle specific operational domains.
Tool Use: We implemented Function Calling to allow the agent to securely interact with external REST APIs.
Backend: The integration layer was built with Node.js and hosted on Cloud Run for seamless scaling.
Challenges we ran into
One major hurdle was ensuring atomic operations—making sure that if a multi-step task failed halfway through, the agent could intelligently roll back or notify a human. We also spent significant time refining our prompts to prevent "hallucinated actions," ensuring the agent only executes commands when it has 100% of the required parameters.
Accomplishments that we're proud of
We successfully achieved a sub-2-second response time for complex tool-triggering tasks. Seeing the agent correctly interpret a vague human request (e.g., "Fix the server lag in the East region") and translate it into a specific set of API calls to scale a Cloud Run service was a huge "Eureka" moment for the team.
What we learned
We learned that the future of AI isn't just about better LLMs, but about better Agentic Orchestration. Understanding how to ground an agent in real-time data and give it "hands" through well-defined tools is a skill set we believe will define the next decade of software engineering.
What's next for Swift Operate
We plan to implement Multi-Agent Collaboration, where a "Manager Agent" in SwiftOperate can delegate specific sub-tasks to specialized "Worker Agents," allowing for even more complex, parallel operational workflows.
Log in or sign up for Devpost to join the conversation.