InsightAgents Multi-agent AI system

Autonomous multi-agent system that extracts insights from your data using Google Cloud and Vertex AI, BigQuery.. Built with ADK to analyze, summarize, and visualize in minutes.

Comment

Inspiration

We were inspired by the growing demand for tools that can understand and reason over unstructured data — such as PDFs, screenshots, URLs, and documents — especially in domains like education, legal, customer support, and research. The introduction of Google’s Agent Development Kit (ADK) allowed us to design intelligent agents that do more than answer questions — they autonomously analyze, derive insights, and collaborate in a cloud-native setup.

What it does

InsightAgents is a multi-agent system designed for data analysis and insights. It consists of:

  • DataCollectorAgent – gathers raw user input or simulated feedback.
  • InsightAnalyzerAgent – uses Vertex AI (Gemini-Pro) to extract actionable insights from the data.
  • ReportGeneratorAgent – compiles the insights into reports and logs them to BigQuery for querying and visualization.

The entire process is automated and orchestrated through a simple pipeline script. It uses Google Cloud's Vertex AI and BigQuery to demonstrate real-world agent collaboration and production-grade analytics.

How we built it

We used Python with the Agent Development Kit (ADK) to define agent logic.

  • Vertex AI's Gemini-Pro model handles natural language analysis and insight generation.
  • We connected BigQuery via the Google Cloud SDK to store insights with timestamps.
  • The .env and service account setup ensures security and reproducibility.
  • The project was built locally using VS Code, authenticated via gcloud, and validated with test scripts for Gemini and BigQuery access.
  • We used Markdown for docs, Lucidchart for the architecture diagram, and ElevenLabs for narration in our demo.

Challenges we ran into

  • Initially, our service account had access issues with Vertex AI’s Gemini model — fixed by manually enabling APIs and adjusting IAM permissions.
  • BigQuery schema setup and JSON insertion had to be carefully structured to avoid type mismatch errors.
  • We had to optimize API calls to fit within the free-tier budget and ensure cost-efficiency.
  • Creating a clean orchestration logic for three different agents without overcomplicating the architecture was a balancing act.

Accomplishments that we're proud of

Successfully implemented an end-to-end multi-agent pipeline with ADK and Google Cloud.

  • Integrated Vertex AI + BigQuery in a way that scales and logs useful analytics.
  • Designed a modular codebase that's easy to extend for future agents.
  • Created a clear and polished architecture diagram and demo video.
  • Delivered a submission that hits all bonus criteria of the hackathon: ADK usage, GCP integration, documentation, and a blog/video walkthrough.

What we learned

  • Deep hands-on experience with Agent Development Kit for building autonomous orchestration logic.
  • How to authenticate and operate Google Cloud services efficiently (Vertex AI, BigQuery).
  • How to deploy and test GenAI-based agents in a real project environment.
  • Valuable insights into coordinating agents in a production-like architecture using local tools and cloud APIs.

What's next for InsightAgents Multi-agent AI system

We plan to:

  • Extend the system with a DocumentProcessorAgent to support PDFs, URLs, and OCR from screenshots.
  • Add Cloud Run deployment for scalable inference.
  • Allow real-time dashboards powered by Looker Studio on top of the BigQuery dataset.
  • Release InsightAgents as an open-source starter kit for researchers, students, and developers interested in multi-agent GenAI pipelines.

Built With

Share this project:

Updates