Process Flow Diagram
Flow Graph and N8N Integration
Multi-browser Agents
Dynamic Summary
Select Workflow

🧠 Beyond Ordinary Browsing (BOB)

“An AI agent that goes beyond simple chat — automating research, visualization, and workflow orchestration across multiple browsers.”

🚀 Inspiration

What inspired us was the need for an intelligent agent that could automate repeated business tasks — from sending routine messages and conducting market research to aggregating financial data.

In the context of the Product Manager challenge, we envisioned a system that could help PMs make faster, more informed decisions by delegating tedious workflows to AI. We wanted to push beyond ordinary browsing, empowering AI not just to read and respond, but to act, reason, and execute across tools and data sources.

🧩 What We Built

Beyond Ordinary Browsing (BOB) is a web-based automation platform that:

Accepts a user prompt describing a business or research task.
Spins up multiple browser agents using browser-use for parallel data gathering.
Uses NVIDIA Nemotron (nvidia/llama-3.3-nemotron-super-49b-v1.5) to plan and reason about what information to fetch.
Summarizes results and visualizes the entire process using an Excalidraw-style flow diagram.
Automatically creates an n8n automation for re-running the same workflow in the future.
Exports tabular data to Google Sheets via Composio for further analysis.

In short, BOB transforms a single user prompt into a repeatable, data-driven workflow.

🧠 What We Learned

Through this project, we learned how to:

Integrate OpenRouter with NVIDIA Nemotron for advanced reasoning and planning.
Use browser-use to spin up parallel agentic browsers, dramatically accelerating data aggregation.
Coordinate multiple specialized agents — one for browsing, one for summarizing, and another for visualization — into a unified orchestration layer.

We also deepened our understanding of agentic architectures, workflow reasoning, and AI-powered productivity tools for product management.

You could think of our approach mathematically as a function composition across agents:

🏗️ How We Built It

We divided responsibilities based on team strengths:

Frontend: Next.js web app with real-time updates and Excalidraw visualization.
Backend: FastAPI backend integrating Nemotron reasoning and browser agent orchestration.
Automation: n8n workflow generation for iterative runs and Composio-based Google Sheets export.
Database & Storage: Convex-based managed session state and user task history for replays.

⚙️ Challenges We Faced

Building BOB pushed us to the limits of both APIs and orchestration:

Browser-use Integration: Implementing and synchronizing multiple browser sessions dynamically required deep exploration of documentation and creative debugging.
Workflow Orchestration: Managing agent coordination, real-time visualization, and n8n workflow generation while keeping performance stable was non-trivial.

Despite these challenges, we successfully built a multi-agent orchestration system that embodies NVIDIA’s goal: enabling AI to take action, not just respond.

🏆 Accomplishments We're Proud Of

Building a fully functional, end-to-end system that integrates Nemotron, browser automation, visualization, and n8n orchestration.
Achieving parallel browser reasoning to gather and synthesize data in real time.
Creating a looping automation pipeline, where workflows can be saved and re-run dynamically.
Pushing the boundary of what “browsing with AI” can mean for real productivity and research by going above simple LLM prompting.

📘 What’s Next for Beyond Ordinary Browsing

Next, we plan to:

Expand support for additional business tools (Email, PDF-style reports, slideshow presentations).
Add team collaboration features, enabling shared research and task orchestration.
Integrate more structured data outputs, like dashboards and analytics summaries.
Explore fine-tuning Nemotron on PM-specific workflows to further align reasoning with real product decision-making contexts.