Microsoft Xbox · Tech Mahindra

AI-powered onboarding: from two weeks to thirty minutes.

When I joined the project, setting up my development environment took nearly two weeks of manual steps, and every new developer went through the same thing. So I documented every missing step, simplified the process, and then built an MCP server with dedicated AI tools for each task. A new developer is now ready in under 30 minutes. I also built tools for the daily project start-up, which used to take 15 to 30 minutes by hand, and up to an hour when someone missed a step. It now runs in under 4 minutes.

My own concept, architecture, and build. Full implementation details shared on request.

All Case StudiesBack to Projects
AI-powered automation pipeline for developer onboarding

Problem

Both first setup and daily start-up were slow and manual.

The project had a complicated setup spread across a 15-step guide and several wiki pages: install the software, configure the web server (IIS), connect the VPN, install and run Docker, download the packages it needs, start the Azure services, set up Visual Studio, and clone the code. It was easy to miss a step, and almost everyone hit problems that were not written down anywhere. Even after setup, starting the project each day meant repeating many of those steps by hand, which took 15 to 30 minutes, and up to an hour when someone missed a step.

Design Decisions

The choices that made it work.

1

Understand the real process first. During my own onboarding I wrote down every step missing from the old docs and built a clear wiki, so I knew the full picture before automating anything.

2

Simplify before automating. I cut steps that were not needed (for example, no longer opening Visual Studio just to build and clean the project over and over), so the tool was not automating busywork.

3

Use the tool developers already have. It runs inside the Copilot chat in VS Code, so there is nothing new to learn.

4

Build it as real, reliable software, not a quick script, with prerequisite checks and error handling so it works on different machines.

5

Automate both halves with dedicated tools: one set for the one-time setup, another for the daily project start-up, since both were manual and repeated.

How It Works

Two chat messages do the whole job.

Once the tool was in place, a developer only needs two messages in the Copilot chat: one to set everything up, and one to run the project. Behind those messages, 21 small AI tools do the work.

1

Message 1: full onboarding setup

The developer opens the Copilot chat and asks for a full setup. The tool takes it from there.

2

Checks access first

It confirms login and access permissions before anything else, so setup never starts on a half-ready machine.

3

Installs and configures everything

It checks prerequisites, installs missing software, sets up the web server, VPN, Docker, certificates, and Azure access, and clones and configures the code, handling errors as it goes.

4

Message 2: run the project locally

A second message triggers the start-up tools: they bring up Docker, start the Azure services, build the solution, and run the container scripts. Tasks that took 15 to 30 minutes by hand every day now finish in under 4 minutes.

5

A daily health check

One command checks network, VPN, access, Docker, addresses, firewall, web server, and certificates, and fixes what it can.

6

The result

Onboarding drops from about two weeks to under 30 minutes, and the daily start-up drops from 15 to 30 minutes of manual steps, sometimes an hour when something was missed, to a single message that finishes in under 4 minutes.

Engineering Challenges

The hard parts, and how I solved them.

The hard part was not calling an AI model. It was the engineering that made the automation reliable on a real, messy work machine.

Two versions of Node.js without clashes

The older part of the app needs an older Node.js, but the AI tool needs a newer one. A small setup script finds or installs the right version on its own, so a new machine just works with no manual config.

Handling changing network addresses

Cloud dev machines get a new network (IP) address on every restart, which breaks the local setup. The tool detects the new address and updates every config file automatically, ending a daily manual chore.

Auto-renewing temporary access

Cloud access permissions expire every 8 hours, and forgetting to renew them breaks deployments. The tool renews them automatically, with no clicking through the Azure portal.

A self-checking daily health check

One command checks network, VPN, access, Docker, address, firewall, web server, certificates, and Node.js, fixes what it can, and clearly reports what still needs attention.

Built With

Real production software, not a demo.

21 tools in TypeScript, built to the MCP standard with progress updates and error handling throughout.

TypeScriptNode.js 20Model Context Protocol (MCP) SDKJSON-RPC 2.0 (stdio)GitHub Copilot Agent ModePowerShell (Windows automation)Azure PIM REST APIAzure CLIWindows IISDocker

Impact

Two weeks of setup became one guided run.

2 wks → 30 min

Onboarding time

30 min → 4 min

Daily start-up

80+ hrs

Saved per developer

What This Shows

From my own pain point to a team-wide tool.

This started with my own onboarding frustration and grew into a tool the whole team uses for both setup and daily work, earning direct appreciation from the Microsoft project lead. It is a good example of how I work: find the real problem, simplify it, then automate it as reliable software.