WorkLeak_EX

"Autonomous Employee Experience: Sensing workplace friction and resolving it before it stalls your team."

Comment

Inspiration

Modern organizations collect huge amounts of workplace data, but identifying that an employee is struggling with a workflow is very different from actually helping resolve the problem.

We were inspired by a simple question:

What if an AI system could detect where employees are losing time, understand why it is happening, and intelligently help resolve the friction?

Existing dashboards can reveal delays, bottlenecks, repeated work, and excessive handoffs, but they often stop at visualization. We wanted to move beyond “Here is the problem” toward “Here is why it happened, what we can do about it, and how we can verify the outcome.”

This led us to build WorkLeak EX — an Employee Experience Intelligence & Action Platform that combines WorkLeak's workflow intelligence with specialized AI agents.

Our core idea became:

Detect → Understand → Decide → Act → Verify → Improve

🚀 What it does

WorkLeak EX detects hidden employee-work friction and transforms those signals into governed, actionable resolutions.

The underlying WorkLeak intelligence identifies patterns such as:

Excessive handoffs Delayed requests Blocked work Repetitive tasks Ownership ambiguity Workflow bottlenecks Long resolution times

WorkLeak EX then uses a multi-agent architecture to investigate and respond to those signals.

🤖 Multi-Agent System

Experience Agent Detects employee friction signals from available workflow data.

Root Cause Agent Investigates evidence and identifies the likely causes behind the friction.

Resolution Agent Creates actionable resolution recommendations.

Governance Agent Evaluates risk and decides whether an action should be allowed, require human approval, or be denied.

Experience Orchestrator Coordinates the complete multi-agent workflow.

🔄 Core Workflow

Employee Problem ↓ Friction Detection ↓ Agent Investigation ↓ Root Cause Analysis ↓ Resolution Recommendation ↓ Governance & Human Approval ↓ Enterprise Tool Action ↓ Verification ↓ Impact Measurement

The platform also provides an explainable Employee Experience (EX) Score, agent execution logs, governance records, and a distinction between projected and measured impact.

The current prototype demonstrates enterprise-tool interaction through a Freshworks Sandbox Adapter and an MCP-ready tool architecture.

🛠️ How we built it

WorkLeak EX was built as an agentic layer on top of the existing WorkLeak intelligence engine.

Architecture WORKLEAK EX │ ▼ EXPERIENCE ORCHESTRATOR │ ┌──────────────┼──────────────┐ ▼ ▼ ▼ EXPERIENCE ROOT CAUSE RESOLUTION AGENT AGENT AGENT └──────────────┼──────────────┘ ▼ GOVERNANCE AGENT │ ┌──────────┴──────────┐ ▼ ▼ ALLOW REQUIRE APPROVAL │ │ │ HUMAN APPROVAL │ │ └──────────┬──────────┘ ▼ TOOL ADAPTER LAYER │ ▼ FRESHWORKS SANDBOX │ ▼ VERIFICATION │ ▼ PROJECTED / MEASURED IMPACT Key technologies and concepts AI / LLM-based reasoning Multi-agent orchestration WorkLeak workflow intelligence Employee Experience analytics Typed enterprise tool adapters MCP-ready architecture Freshworks Sandbox integration Human-in-the-loop governance Deterministic EX scoring Audit and execution logging Structured AI outputs Projected vs measured impact modeling Modern web application architecture

We intentionally separated implemented capabilities, sandbox functionality, projected results, and future roadmap items so that the prototype remains technically transparent.

⚡ Challenges we ran into

Building an agentic system was more challenging than simply connecting an LLM to a dashboard.

  1. Turning analytics into actions

WorkLeak could identify workflow friction, but we needed to determine how an agent could safely move from:

“We found a problem”

to:

“Here is an appropriate action.”

We solved this by introducing specialized agents with clearly defined responsibilities.

  1. Preventing uncontrolled AI actions

An autonomous agent should not automatically execute every action it generates.

We introduced a governance layer:

ALLOW → REQUIRE APPROVAL → DENY

This gives the system a controlled path from AI recommendation to execution.

  1. Separating projected and measured impact

It was tempting to immediately show an estimated improvement after an action.

Instead, we separated:

Projected Impact

from:

Measured Impact

so the platform does not present an expected result as a verified result.

  1. Enterprise integration complexity

Enterprise platforms have authentication, permissions, API limitations, and different tool interfaces.

Rather than creating fake production integrations, we designed a typed adapter architecture with a Freshworks Sandbox Adapter and an MCP-ready tool layer.

  1. Making multi-agent reasoning explainable

We wanted judges and users to understand why an agent made a recommendation.

Therefore, agent runs, tool calls, evidence, governance decisions, and outcomes are represented through execution and audit information.

🏆 Accomplishments that we're proud of

We are proud that WorkLeak EX goes beyond being another AI chatbot or analytics dashboard.

🤖 A genuine multi-agent workflow

Different agents have different responsibilities instead of using one general-purpose AI prompt for everything.

🛡️ Governed agentic automation

AI recommendations pass through a governance layer before actions can be executed.

📊 Explainable EX Score

The Employee Experience Score uses deterministic components rather than allowing an LLM to arbitrarily generate a score.

🔍 From detection to action

WorkLeak's existing intelligence is connected to an agentic decision-making layer.

🔄 Projected vs measured impact

The platform explicitly distinguishes expected outcomes from evidence-based measurements.

🔗 Enterprise-ready architecture

The typed tool-adapter approach provides a foundation for connecting enterprise systems while keeping sandbox and production capabilities separate.

🎯 Strong hackathon alignment

The project directly addresses Track 1 — Customer & Employee Experience through AI agents, orchestration, enterprise tools, governance, and employee workflow intelligence.

📚 What we learned

Building WorkLeak EX taught us that creating an agentic system is not simply about making an LLM autonomous.

We learned that a useful enterprise agent needs:

Context → Tools → Reasoning → Governance → Execution → Verification

We also learned the importance of:

Giving each agent a clear responsibility Validating AI-generated outputs Using structured tool contracts Keeping humans involved in high-risk decisions Recording agent and tool activity Separating simulated and real integrations Never presenting projected results as measured results Designing for failure instead of assuming every tool call succeeds

Most importantly, we learned that trust is as important as intelligence when building AI systems that can influence real workflows.

🔮 What's next for WorkLeak EX

The current prototype establishes the foundation for a broader Employee Experience intelligence platform.

Next steps include:

  1. Live Freshworks Integration Connect the server-side Freshworks adapter to real enterprise workflows.

  2. Real MCP Server & Client Integration Extend the MCP-ready tool architecture into a verified MCP implementation.

  3. Continuous Experience Monitoring Monitor employee workflow signals continuously rather than relying primarily on existing datasets.

  4. Real Post-Action Measurement Compare actual post-action results against baseline metrics to produce verified impact measurements.

  5. More Enterprise Systems Expand the adapter architecture toward systems such as Jira, GitHub, Slack, Microsoft 365, and other workplace platforms.

  6. Personalized Employee Experience Intelligence Move from organization-level insights toward context-aware workflow assistance while maintaining privacy and governance.

  7. Predictive Friction Detection Identify potential workflow bottlenecks before they significantly affect employees.

  8. Customer Experience Expansion Extend the same intelligence architecture from Employee Experience toward Customer Experience workflows.

Built With

Share this project:

Updates