From AI Curiosity to a Working AI Service Desk in 7 Days
How a leading global technology services company upskilled its business operations teams with our AI Level-2 Deep Dive for BizOps, and what happened when every participant built a real, working AI system.
Every operations leader we speak to is asking the same question: how do we move our teams from “we should be using AI” to “we are using AI, safely, every day”?
This is the story of a program that answered that question in seven days. In September and October 2026, we delivered the second batch of our AI Level-2 Deep Dive for BizOps for a leading global technology services company. It was a repeat engagement with the same client, and it delivered a 4.7 out of 5 overall rating, a room of engaged participants, and a working AI Service Desk Orchestrator built by the learners themselves.
The challenge: AI awareness is not AI capability
Business operations teams sit at the centre of how enterprises run. They manage tickets, SLAs, SOPs, escalations, reports and vendor workflows, and they are exactly where generative AI can remove the most repetitive work. Yet most AI training aimed at these teams stops at awareness: a few demos, some prompt tips, and a certificate.
Our client wanted something different. They needed their teams to:
- Understand how LLMs, RAG and agents actually work, in plain business language
- Write reliable, repeatable prompts for real Ops tasks such as RCA, SOP extraction and escalation summaries
- Automate recurring workflows without waiting on engineering
- Build something tangible, and govern it responsibly
Our approach: build while you learn
We designed the program around one principle: every concept should end in something participants have built or run themselves. Three design choices shaped everything.
1. A capstone woven through the curriculum
Instead of a separate project at the end, the capstone, an Intelligent Service Desk Orchestrator (ISDO), is introduced on Day 5 and built lab by lab across the weekend. Every lab adds a working component, so learning and building happen together.
2. Claude Desktop as the AI development partner
Labs follow a “describe intent, generate, run and validate” pattern. Participants tell Claude Desktop what they want, review the code it generates, and test it. This lowers the barrier for non-developers while still teaching the thinking behind the system. It also helped professionals from non-IT backgrounds get started with confidence.
3. Responsible AI from day one
PII handling, bias, audit trails and human approval gates were not an afterthought. They were applied directly to the system participants were building.
The 7-day journey
The program ran as five weekday sessions of 2.5 hours, followed by two intensive 6-hour weekend sessions.
| Day | Focus | What participants did |
|---|---|---|
| Day 1 | Applied AI Foundations | Compared RAG and non-RAG answers on a real SOP corpus; tested how temperature, constraints and few-shot examples change output |
| Day 2 | Advanced Prompt Engineering | Applied the TFCE framework (Task, Format, Constraints, Examples) to build RCA prompts, SOP extractors and an executive status-report summariser |
| Day 3 | AI Tools for BizOps | Explored AI in M365 Copilot, JIRA, ServiceNow and Confluence; automated inbox replies with tone and policy controls; generated KPI narratives |
| Day 4 | No-Code / Low-Code Automation | Designed email-to-ticket flows and weekly report automation across Power Automate, Zapier, Make.com and n8n |
| Day 5 | SLA Alerts + Capstone Preview | Built SLA breach triggers, walked through the full ISDO architecture and set up the project scaffold in git |
| Day 6 | GenAI Data Skills, Agents, Responsible AI | Built a ChromaDB knowledge base, mock ServiceNow and Jira APIs, a Triage Agent and a Resolution Agent; created a Responsible AI checklist |
| Day 7 | Full Capstone Build and Showcase | Added SLA and escalation logic, LangGraph orchestration, human-in-the-loop approvals, an A2A knowledge specialist, PII redaction and live demos |
The capstone: an AI-powered IT Service Desk Orchestrator
The Day 7 deliverable was a multi-agent system that takes a ticket from intake to resolution or escalation. The target was to automate 60 to 70 percent of the repeatable ticket workflow while keeping humans in control of risky decisions.
Triage Agent
Reads each ticket and assigns category, priority and assignment group, tested against held-out tickets.
Resolution Agent
Searches a ChromaDB knowledge base and suggests a fix, with low-confidence matches routed for human review.
SLA & Escalation Agent
Detects breach risk early and triggers P1 escalation.
Supervisor (LangGraph)
Orchestrates the full flow with conditional routing, retries and fallbacks.
Human-in-the-Loop Gate
Pauses for approval on access grants, P1 escalations and uncertain answers, then resumes cleanly.
Knowledge Specialist (A2A)
A standalone FastAPI service the Resolution Agent calls for complex queries, connected through a custom MCP server.
On top of this sat a PII redaction layer and a full audit trail: names, emails, employee IDs and IP addresses masked in agent logs, and every action recorded with the agent, tool, rationale and approval status.
The program closed with each participant running their own system against an evaluation ticket set in a live showcase, followed by a discussion on what it would take to move the proof of concept to a production pilot.
The results
We collected feedback from participants right after the final session. The numbers (out of 5), from 30 responses:
In participants’ own words
What made it work
- Hands-on from the first hour. Theory was always followed by a lab, and every lab fed the capstone.
- Real BizOps scenarios. Tickets, SLAs, SOPs and escalations kept every exercise relevant to participants’ day jobs.
- Governance built in, not bolted on. Participants learned to ask “is this safe to deploy?” while building, not after.
- A patient, energetic trainer. Maruti’s clarity and willingness to walk through every step were mentioned again and again in the feedback, including by learners from non-IT backgrounds.
- A strong partnership. The client’s support and coordination, together with our own organising team, kept a 7-day live program running smoothly.
Takeaways for L&D and business leaders
- Pick a real deliverable. A working proof of concept changes how teams talk about AI internally.
- Start with the people closest to the process. Ops and service delivery teams know where the friction is, and with the right guidance they can prototype solutions themselves.
- Treat Responsible AI as a skill, not a policy. Teams that practise redaction, approvals and audit logging while building carry those habits into production.
- Plan the next step before the program ends. The showcase conversation about moving to a production pilot is where training turns into transformation.
Continue the journey: related programs
If this story resonates, these programs build on the same skills and can be tailored for your teams:
Want to get a feel for agent building first? Join our upcoming webinar, Building Your First Agent with LangGraph (8 October 2026), or browse more training success stories and our AI training programs.
A heartfelt thank you
To our client, thank you for trusting us with this opportunity, and for investing in hands-on AI capability for your people. Delivering a second batch for you is a privilege we do not take lightly.
To every participant, thank you for the energy, curiosity and commitment you brought across all seven days. You made this batch what it was.
And to the Optimistik Infosystems organising team, thank you for the planning, coordination and behind-the-scenes effort that kept every session on track. Great training looks effortless because of people like you.
Ready to see what your team could build?
Whether you lead Operations, Service Delivery, PMO or L&D, we can design a hands-on GenAI program around your tools, your workflows and your goals. Your team’s first working AI system could be a week away.
Explore the Program Talk to Our Team
