Inspiration

The idea for klemarKlemer came from a problem we frequently encounter in insurance claims operations: claims can get stuck because tasks are not always assigned to the right person at the right time.

In our experience, a claim can be delayed when an employee receives more tasks than they can handle, while another employee with the right skills and available capacity may have fewer tasks.

This becomes even more challenging for surveyors, where workload, availability, current task status, and operational conditions can affect how quickly they can complete an assignment.

We wanted to solve this problem by creating a system that can understand the current workload of each employee and determine who should handle each task.

Instead of using a simple round-robin assignment or always giving work to the highest-performing employee, klemarKlemer uses Gemini to make task assignment decisions based on the available operational information.

Our goal is to make claim operations more proactive: give the right task to the right person at the right time, while continuously monitoring the workload and performance of the team.

What it does

klemarKlemer is an agentic claims operations platform focused on intelligent task assignment and SLA management.

The core of the system is an AI-powered assignment process.

Intelligent Task Assignment

When a new task needs to be assigned, klemarKlemer collects information about the available employees, including:

Current workload.
Employee skills.
Claim type experience.
Current task status.
Historical SLA performance.
Task complexity.
Availability.
Surveyor specialty when the task requires a survey.

This information is provided to Gemini, which determines the most suitable employee for the task.

For example:

New Claim Task
      |
      v
Employee Information
      |
      +-- Skills
      +-- Workload
      +-- Experience
      +-- SLA Performance
      +-- Availability
      +-- Current Tasks
      |
      v
     Gemini
      |
      v
Recommended Employee

The goal is not simply to choose the employee with the highest performance score.

Instead, Gemini can consider the current situation.

For example, an employee might have excellent historical performance but already have several urgent claims. Another employee may have slightly lower historical performance but the correct skills and enough capacity to handle the new task.

The second employee may therefore be the better choice.

Surveyor Assignment & Live Tracking

For survey-related claims, klemarKlemer can also track the survey task through its operational lifecycle:

Survey Requested
      ↓
Assigned
      ↓
Accepted
      ↓
En Route
      ↓
Arrived
      ↓
Surveying
      ↓
Completed

This provides operations teams with visibility into the current state of survey tasks.

The current operational state can also become input for future assignment decisions.

For example, if a surveyor is already handling several active surveys, Gemini can take that workload into consideration when deciding who should receive the next survey.

SLA Monitoring

The platform monitors tasks and claims against their SLA.

This allows the system to identify:

Tasks approaching their SLA.
Overdue tasks.
Employees with excessive workloads.
Survey tasks that are not progressing.
Claims waiting too long for assignment.

This helps the operations team react before a small delay becomes a larger SLA problem.

Performance Feedback

Performance information is not only used for reporting.

It can also become part of the assignment decision.

The system can track metrics such as:

SLA compliance.
Average completion time.
Number of completed tasks.
Current workload.
Rework rate.
Task response time.

This creates a continuous feedback loop:

Assignment Decision
        ↓
Task Execution
        ↓
Live Operational Data
        ↓
Performance Data
        ↓
Future Assignment Decision

This means the system can continuously improve how work is distributed across the team.

How we built it

We built klemarKlemer as a full-stack application with an agentic assignment layer.

Frontend

React
Vite

The frontend provides the claims operations console where users can view claims, employees, assignments, task status, and operational information.

Backend

Go
Candi Clean Architecture
PostgreSQL
Redis

The backend manages claims, employees, tasks, assignments, and operational events.

AI

Gemini 3.5 Flash
Vertex AI / Gemini API

Gemini is primarily used as the decision-making engine for task assignment.

The backend gathers the relevant operational context and provides it to Gemini so that the model can determine which employee is the most appropriate person for the task.

The system therefore combines structured business data with AI reasoning:

Claims Data
     +
Employee Data
     +
Workload
     +
Performance
     +
Availability
     +
Task Requirements
     |
     v
   Gemini
     |
     v
Assignment Decision

We also maintain the underlying operational event history so that assignment decisions and task progression can be traced.

Challenges we ran into

The biggest challenge was figuring out how to make AI assignment useful instead of simply asking Gemini to pick a random employee.

We needed to give the model enough context to make a meaningful decision.

For example:

Employee A
Skill: Excellent
Current Workload: High
SLA Performance: Excellent

Employee B
Skill: Good
Current Workload: Low
SLA Performance: Good

Choosing Employee A simply because they have better historical performance could actually make the overall operation worse.

The system therefore needs to consider multiple factors at the same time.

Another challenge was making the assignment decision reflect the current state of the operation.

Employee availability and workload can change constantly. A decision that was correct an hour ago may no longer be correct now.

This is why live task status and workload information are important inputs to the assignment process.

For surveyors, this becomes even more challenging because their tasks happen in the field. Their current task, availability, and operational status can affect whether they are realistically able to accept another survey.

We also had to think carefully about human control. Gemini recommends the assignment, but the system should still provide operational users with visibility into the decision and allow appropriate human intervention.

Accomplishments that we're proud of

We are proud that klemarKlemer focuses on a real operational problem rather than simply adding AI to an existing workflow.

Our key accomplishments include:

Built an AI-powered task assignment system using Gemini.
Used employee workload and skills as context for assignment decisions.
Considered historical performance together with current workload.
Built a surveyor-specific workflow with task status tracking.
Created operational visibility into where tasks are currently stuck.
Connected performance information to future assignment decisions.
Implemented SLA monitoring around claims and tasks.
Created an event history for tracking operational changes.
Built the system so the AI decision is part of an actual business workflow rather than an isolated chatbot.

The part we are most proud of is the feedback loop:

        New Task
           ↓
    Gemini Assignment
           ↓
      Employee
           ↓
     Task Execution
           ↓
   Live Task Tracking
           ↓
 Performance & Workload
           ↓
    Gemini Assignment

The system doesn't just answer "Who is the best employee?"

It tries to answer:

"Who is the best person to handle this task right now?"

What we learned

We learned that an agentic system is only useful when the agent has access to the right context.

For task assignment, simply knowing an employee's historical performance is not enough.

The AI needs to understand the combination of:

skill + workload + availability + experience + task requirements + performance.

We also learned that the highest-performing employee is not necessarily the right person for the next task.

Optimizing for individual performance without considering workload can create bottlenecks.

Another important lesson was the value of live operational data.

For surveyors especially, knowing whether someone is available, already assigned to another survey, or currently working on a task can significantly change the assignment decision.

Finally, we learned that AI should support operational decision-making rather than operate as a black box. The system should provide the information behind an assignment and allow humans to understand and intervene when necessary.

What's next for klemarKlemer

Our next goal is to make the assignment engine more intelligent and operationally useful.

Smarter Workload Balancing

We want Gemini to consider workload across the entire team and proactively identify potential bottlenecks before assigning new tasks.

Real-Time Surveyor Operations

We want to improve the surveyor dashboard with real-time operational states:

Available
Assigned
Accepted
En Route
Arrived
Surveying
Completed
At Risk
Overdue

This would allow operations teams to immediately see which surveys need attention.

Location-Aware Assignment

For field surveys, we want to explore using appropriate location information to help determine which available surveyor can reach a location efficiently, while following the company's privacy and data governance policies.

Predictive SLA Risk

Instead of only detecting an SLA breach after it happens, we want the system to identify tasks that are likely to miss their SLA and take action earlier.

For example:

Task At Risk
     ↓
Check Employee Workload
     ↓
Check Task Progress
     ↓
Check SLA Remaining
     ↓
Re-evaluate Assignment
     ↓
Notify / Escalate / Reassign

Continuous Assignment Optimization

Ultimately, we want klemarKlemer to continuously learn from operational outcomes.

Task
 ↓
Assignment
 ↓
Execution
 ↓
Performance
 ↓
Outcome
 ↓
Better Assignment

Our long-term vision is to transform claims operations from a reactive process where teams discover stuck claims after the SLA is already at risk into a proactive operation where every task is continuously routed to the right person based on the team's current situation.