TEAM TENSOR

• 

  Resolution Memory Every successful outcome becomes reusable evidence, helping the swarm resolve similar incidents faster and safely.

• 

  Parallel Agent Investigation Specialized AI agents investigate payments, orders, and customer context simultaneously.

• 

  Customer Resolution Experience Customers receive a clear explanation and choose their preferred resolution—without contacting support.

• 

  Order Restored ResolveX safely restores the missing order using the original payment, with no duplicate charge.

• 

  1. Operations Dashboard ResolveX Swarm proactively detects customer-impacting failures before they become support tickets.

• 

  Root Cause and Safety Decision The swarm identifies the root cause, evaluates safety policies, and recommends a verified resolution.

Inspiration

Normal customer support usually starts only after something breaks. A customer sees a problem reaches out for help tells their story over and over. Then waits while someone looks through many different systems.

We wanted to change that.

ResolveX Swarm came from one idea:

What if an AI team could find a problem look into it and fix it safely before the customer even knows they need help?

We looked at an annoying problem: a payment goes through but the order is never made. The customer sees money leave their bank account but gets no order confirmation. They have no way of knowing if their money is safe.

ResolveX Swarm turns this support into a proactive workflow:

[

\text{Predict} \rightarrow \text{Detect} \rightarrow \text{Investigate} \rightarrow \rightarrow \text{Learn}

]What it does

ResolveX Swarm finds mismatches across company systems and uses special AI agents to look into them.

In our demo ResolveX Swarm sees that 23 customers paid for things but have no orders. Then ResolveX Swarm starts three investigations at the same time:

• The Payment Agent checks the transaction.

• The Order Agent looks to see if an order exists.

• The Customer Context Agent finds customer details and past support notes.

The Root Cause Agent puts all the facts together while the Resolution Agent figures out how to fix it. Finally the Safety Guardian checks the risks, rules and permissions.

Depending on the risk ResolveX Swarm can:

• Fix low-risk problems automatically.

• Ask the customer to confirm the fix.

• Ask an employee to approve the fix.

For the ₹4,999 demo transaction the customer can fix the order ask for a refund or talk to a person. If they choose to fix the order ResolveX Swarm connects the payment and makes sure they are not charged again.

After that the result is saved in resolution memory. If a similar problem happens later the system can use that info but it always runs safety checks again.

How we built it

ResolveX Swarm is a web app built with React, TypeScript, Vinext and Vite.

We split the system into layers:

• Agent logic

• Incident orchestration

• Enterprise tool interfaces

• Mock enterprise services

• Customer and employee experiences

• Resolution memory

• Presentation components

The integration layer uses an MCP-style interface. It uses tools to find customers, transactions orders and tickets. It can also restore orders give refunds tell customers what is happening and send cases to humans.

Since the tools use an interface we can swap our tools for real enterprise APIs later without having to rewrite the agents or the user interface.

Investigations run at the time:

await Promise.all([

paymentAgent.investigate()

orderAgent.investigate()

customerContextAgent.investigate()

]);

The interface shows updates as each agent finishes its work. We do not show the internal thoughts of the AI. Instead we show summaries like "Payment verified," "Order record missing " and "Policy check passed.”

The final prototype includes:

• A live operations dashboard

• Animated multi-agent investigation

• Root-cause and confidence reporting

• Risk-based resolution controls

• A calming customer experience

• An employee approval panel

• Simulated order restoration

• Structured resolution memory

• A automated demo mode

Challenges we faced

Making the system feel different from a chatbot

Our biggest challenge was not making this look like a standard chat box. We wanted ResolveX Swarm to feel like a working system rather than just a chatbot waiting for questions.

We did this by showing the incidents the evidence, the agents and the actions as a workflow.

Showing agent activity without showing thinking

We wanted people to see what the agents were doing without showing every technical detail or private reasoning.

We kept the processing separate from the simple summaries. The interface shows facts, progress and decisions.

Balancing automation with safety

Fixing everything would be fast but it could be dangerous. Asking a human for help every single time would make the system too slow.

We made three risk levels:

1. Low risk: fix it automatically.

2. Medium risk: ask the customer first.

3. High risk: ask an employee to approve it.

This lets the platform work fast while still letting humans stay in control.

Coordinating work

Checking payments orders and customer info can all happen at once. Doing them one by one would take long and feel slow.

We set them up to run at the time so the agents can work together quickly.

Making tools feel real

A simple screen would not show how the product actually works. So we built services that act like real systems with realistic wait times. When an order is restored the system actually updates the order links the payment and saves the result.

What we learned

We learned that people trust -agent systems much more when they can see:

• Which systems were checked

• What facts were found

• What actions can be taken

• Why they need to approve something

• If the action worked

We also learned that regular code is often better than AI for tasks. Checking a transaction status or a permission is faster and cheaper with logic.

AI agents are best at putting evidence finding patterns and talking clearly—not for replacing every single business rule.

Finally we learned that how the customer feels is just as important as the tech. Telling a customer "Your money is safe" is much better than showing them an error code.

Accomplishments that we are proud of

We are proud that ResolveX Swarm is more than a static dashboard. The whole process works from start to finish:

• An incident is found.

• Agents look into it at the time.

• A cause is found.

• Safety and policy checks run.

• The customer is asked to confirm.

• The order is fixed using a tool.

• The result is. Saved.

We are also proud of how we organized the code. This makes it easy to turn this prototype into a tool for big companies.

What’s next

The next step is to connect ResolveX Swarm to payment and order systems using MCP servers.

Future versions could add:

• Real-time event tracking

• ways to find problems

• Custom business rules

• Logs and history of approvals

• Sending cases, to human experts

• Data on how well agents work

• Constant testing of agent accuracy

• Secure company logins and permissions

• More problems to fix, like failed refunds or shipping delays

Our long-term goal is to make customer support something that stops problems before they even happen.

Built With

• agent
• agents
• ai
• apis
• asynchronous
• automation
• cloudflare
• enterprise
• mcp
• multi-agent
• openai
• orchestration
• parallel
• processing
• react
• resolution
• risk-based
• services
• sites
• systems
• typescript
• vinext
• vite
• workers