Operator

Give AI a business goal. Let it plan, research, analyze, qualify, act, and review the work.

Comment

Inspiration

The idea for Operator came from a simple problem we kept seeing in businesses.

You can use AI to write an email, find information, analyze something, or generate a plan. But in most cases, you still have to tell the AI what to do at every step.

We wanted to try something different.

What if a business owner could simply say:

"Find 10 high-value healthcare businesses in India that need a new website."

…and the AI could figure out the steps itself?

That became the idea behind Operator.

Instead of building another chatbot, we wanted to build something that feels more like a digital business operator.

What Operator does

You give Operator a business goal.

It then works through the goal step by step:

GOAL → PLAN → RESEARCH → ANALYZE → QUALIFY → ACT → REVIEW → RESULT

For example, if the goal is to find potential customers, Operator can plan the research, look at the available information, identify useful signals, score the opportunities, prepare personalized outreach, and review the result before showing it to the user.

The user doesn't have to manually create every step.

At the same time, we didn't want the AI to blindly take actions on someone's behalf. Important external actions stay behind human approval.

How we built it

We built Operator as a multi-agent system.

Different parts of the work are handled by different agents:

  • Planner figures out what needs to be done.
  • Researcher collects information and evidence.
  • Analyst looks for useful signals in that information.
  • Qualifier scores and ranks the opportunities.
  • Writer prepares personalized outreach.
  • Reviewer checks the generated work before it reaches the user.
  • Operator coordinates everything and keeps track of the mission.

The frontend is built with React and the backend uses FastAPI.

We also keep the mission state, leads, research, scores, outreach, follow-ups, and activity history so the user can see what happened instead of receiving one unexplained AI response.

Nebius + NVIDIA

Nebius was an important part of the project because we wanted the actual agent workflow to use real inference rather than just presenting a simulated AI experience.

Operator connects its AI layer to Nebius Token Factory, where we use an NVIDIA open-source model for the AI tasks in the workflow.

The basic flow is:

Operator → Agent Orchestrator → AI Provider → Nebius Token Factory → NVIDIA Model

We kept the AI provider separate from the rest of the application so that the agents and workflow don't have to be rewritten when the inference provider changes.

What makes Operator different

The main idea is actually quite simple:

Most AI tools ask: "What should I do?"

Operator asks: "What are you trying to achieve?"

That's the part we wanted to explore.

The user gives the outcome, and Operator works backwards to figure out the tasks needed to get there.

It isn't meant to completely remove humans from the process. Instead, it takes care of the repetitive parts and gives the user enough information to understand and approve the important actions.

Challenges

The hardest part was not getting an AI model to generate text.

The harder problem was making the whole thing behave like one product.

If seven different agents are involved, the user shouldn't feel like they are using seven different AI tools. The workflow has to feel like one continuous process.

We also wanted users to be able to see what was happening. That's why the mission view shows the different stages, agent activity, research, scores, and review results.

Another challenge was keeping the system usable in both demo and live inference environments while keeping the actual AI integration separate from the application logic.

What we learned

While building Operator, we realized that an AI agent becomes much more useful when it can do more than generate an answer.

It needs to be able to plan, gather information, make decisions, check its own work, and keep the user in control.

That's what we tried to build with Operator.

The idea we want to take forward is simple:

Give AI the goal. Let it figure out the work.

Built With

Share this project:

Updates

Submission history