Project: Enterprise Agents - SmartDesk AI — Enterprise IT HelpDesk Command Center

Track

Enterprise Agents (M365 Agents Toolkit)

Project Name

SmartDesk AI — Enterprise IT HelpDesk Command Center

GitHub Username

@tsechis

Repository URL

https://github.com/tsechis/SmartDesk-AI

Project Description

SmartDesk AI

SmartDesk AI is a connected multi-agent Enterprise IT HelpDesk built in Microsoft Copilot Studio and used directly in Microsoft 365 Copilot Chat.

The project solves a common enterprise problem: IT support is often fragmented across tools, slow triage, repeated user explanations, and unclear ownership. SmartDesk AI centralizes these workflows into one conversational entry point, then intelligently routes each request to the right specialist agent.

What the project does

It helps employees:
troubleshoot technical issues,
create and manage IT tickets,
request software and access,
get policy/compliance guidance,
improve cybersecurity awareness.

What it solves

Reduces time to resolution through automatic intent routing.
Eliminates repeated context by passing structured summaries between agents.
Improves support consistency with guided workflows and policy-aware responses.
Brings all helpdesk interactions into the Microsoft 365 ecosystem (Teams/Copilot Chat).

Main features

Intelligent orchestration of 5 specialist generative agents.
Guided troubleshooting for hardware, software, network, email, printer, and collaboration issues.
Ticket lifecycle management (create, track, escalate, close) with Dataverse MCP integration.
Software & access provisioning with catalog and approval flow support.
IT policy advisor for password, BYOD, acceptable use, and compliance questions.
Security awareness coaching with interactive phishing quizzes and best practices.
Adaptive Cards UX for structured, interactive user flows.

Demo Video or Screenshots

Screenshots: https://github.com/tsechis/SmartDesk-AI/tree/main/screenshots

Primary Programming Language

Other

Key Technologies Used

• Microsoft Copilot Studio (multi-agent orchestration, generative agents)
• Microsoft 365 Copilot Chat (end-user runtime interface)
• Connected Agents architecture (Orchestrator + 5 specialist agents)
• Microsoft Dataverse (persistent ticket and request data)
• Dataverse MCP Server (CRUD operations for tickets/provisioning)
• OAuth 2.0 + Microsoft Entra ID (authentication and authorization)
• Adaptive Cards (interactive conversational UI)
• Microsoft Teams / Microsoft 365 ecosystem (delivery channel)
• Power Platform Solution packaging (deployment/import)

Submission Type

Individual

Team Members

No response

Submission Requirements

• My project meets the track-specific challenge requirements
• My repository includes a comprehensive README.md with setup instructions
• My code does not contain hardcoded API keys or secrets
• I have included demo materials (video or screenshots)
• My project is my own work with proper attribution for any third-party code
• I agree to the Code of Conduct
• I have read and agree to the Disclaimer
• My submission does NOT contain any confidential, proprietary, or sensitive information
• I confirm I have the rights to submit this content and grant the necessary licenses

Quick Setup Summary

1. Clone the repo
2. Import solution in Microsoft Powerplatform (solution is unmanaged) SmartDeskAIEnterpriseITHelpDeskCommandCenter.zip
3. Upload knowledge bases from the knowledge/ folder to each agent:
   • SmartDesk → helpdesk-orchestrator-guide.md
   • IT Troubleshooter → troubleshooting-guide.md
   • Ticket Manager → helpdesk-ticket-playbook.md
   • Access & Software Provisioner → software-access-catalog.md
   • IT Policy Advisor → it-policies-reference.md
   • Security Coach → security-awareness-training.md
4. Publish Ticket Manager agent
5. Publish Access & Software Provisioner agent
6. Publish SmartDesk agent

Technical Highlights

1. Duplicate-response prevention in multi-agent flows
   • Solved with structural routing instructions
2. Hybrid orchestration for control + flexibility
   • Kept the Orchestrator structurally constrained, while all specialist agents remained fully generative for scalable domain handling.
3. Reliable Dataverse MCP schema alignment
   • Enforced describe_table before first create in each session, with mismatch/error fallbacks to avoid invalid writes.
4. Structured cross-agent context handoff
   • Introduced hidden summary payloads (e.g., diagnostic/provisioning/policy summaries) so users never need to repeat information.
5. Robust Adaptive Card event handling in Teams - Standardized submit handling (value.action / msteams.messageBack) to prevent double-submit and duplicate processing.

Challenges & Learnings

Challenges Faced

1. Duplicate Message Prevention: The biggest UX challenge with multi-agent setups in Teams — after a connected agent responds, the Orchestrator's generative layer can produce an additional message (JSON metadata log or echo). Solved with 5 structural routing topics (Route_Troubleshooter, etc.) where each topic contains: Trigger → Connected Agent node → "End current topic", giving the generative layer no opportunity to produce output.

2. Hybrid vs. Pure Generative Architecture: Pure generative orchestration (no Topics) gave the AI too much freedom to produce unwanted post-agent output. A single routing topic in the Orchestrator provides structural control while keeping all 5 connected agents fully generative.

3. MCP Schema Mapping: Ensuring agents correctly map input to Dataverse logical names required mandatory describe_table calls before first create in each session, with failure handling for schema mismatches.

4. Cross-Agent Context Passing: Designing structured summary formats (DIAGNOSTIC SUMMARY, PROVISIONING SUMMARY, etc.) that carry enough context for seamless handoffs while remaining invisible to users.

5. Adaptive Card Interaction in Teams: Handling card submit events correctly (value.action vs plain text) to prevent double-processing and duplicate replies.

Key Learnings

1. Structural > Instructional: Instruction-level rules alone cannot prevent the generative orchestrator from producing unwanted output. Structural fixes (routing topic with "End current topic" after each Connected Agent node) are essential.

2. Generative Agents Scale Well: Once the instruction + knowledge pattern is established, adding new specialist agents is straightforward — no topic flow maintenance required.

3. Knowledge Base Quality Matters: The quality and structure of knowledge base documents (decision trees, catalogs, policy tables) directly determines agent response quality.

4. Read-After-Write is Essential: Never trust a Dataverse write operation without verification — always confirm with a read-back before claiming success to the user.

5. Connected Agent Pattern: The Copilot Studio connected agents architecture enables natural division of responsibility and clean separation of concerns.

Contact Information

luk.oplt@gmail.com

Country/Region

Czech Republic

This submission describes the SmartDesk AI project rather than a scoped change for contributors. Start by reviewing README.md, the knowledge/ files, and the packaged solution referenced in the setup steps; no specific implementation target or completion criterion is provided.