The demo works…

Then a provider goes down, a tool call refunds the wrong customer, a cache serves “cancel” to someone who wrote “do not cancel”, or a crash wipes half a workflow.

…these projects handle what comes after.

Each one takes a production failure, fixes it in .NET 10, and proves the fix with tests you can run on your own machine.

What you get

Inside the membership.

A complete .NET 10 project
Full source that builds and runs, ready to study, clone and adapt.
Tests you can run
Every project ships with its test suite, so you can see each fix hold up.
A step-by-step PDF guide
It explains every part of the project and the decisions behind it.
The whole library
You get immediate access to everything published so far.

Project Gallery

These are just a few highlights — the library has 20+ complete projects, and a new one drops every day.

Each highlight shows the number of automated tests it ships with.

Two AI providers with a router switching traffic to the healthy one — **Resilient AI failover**When one AI provider goes down, your app keeps answering.48 automated tests

An AI agent holding a scoped key in front of a locked MCP server — **Secure MCP server**Give your AI agent scoped access, never the master key.52 automated tests

A chatbot filtering documents by each user's permissions — **Permission-aware RAG**A RAG chatbot that only answers with what each user is allowed to see.82 automated tests

An invoice turned into validated structured data — **Reliable invoice extraction**Valid JSON isn't enough — we validate the numbers too.120 automated tests

A CI pipeline with an evaluation gate checking an AI agent — **Agent evals in CI**Your agent passed the demo; now it passes CI, every commit.141 automated tests

A chat answer stream resuming after a dropped connection — **Resumable streaming**Wi-Fi blinks, the answer picks up exactly where it left off.108 automated tests

An AI agent waiting for a person to approve a refund — **Human approval for agents**Your agent asks before it refunds a customer.56 automated tests

A semantic cache blocking a near-match prompt with opposite meaning — **Semantic caching**Cut LLM costs without serving the wrong "close enough" answer.97 automated tests

A workflow resuming from a saved checkpoint after a crash — **Durable AI workflows**Crash mid-workflow, resume from the last checkpoint — nothing lost.92 automated tests

Get all 20+ projects for $5/month

“Wrap only the dangerous tools in ApprovalRequiredAIFunction. Lookups stay fast and auto-run; refunds and cancels wait for a person.”

From our post on human approval for AI agents in .NET 10. Read the article

Read before you join

Two recent free articles on Medium. The full project behind each one is in the membership.

Semantic Caching for LLM Calls in .NET 10: When “Close Enough” Is a Bug

A plain similarity threshold can replay the answer to “cancel my order” when someone asks “do not cancel my order”. This article builds a semantic cache on Microsoft.Extensions.AI that serves an earlier answer only when it’s safe. It runs offline with fake models, so your CI never needs an API key.

Plugging MCP Servers into a .NET 10 Chat App Without Trusting Them

Real MCP servers bring dangerous tools, name clashes, slow calls and results that try to instruct your model. This article builds a chat host on the official MCP C# SDK with allowlists, prefixed tool names, timeouts, truncation and fenced results, and it keeps the chat working when a server is down.

Who it’s for

.NET developers adding AI features in C#
Junior and mid-level developers who want senior-level patterns
Anyone preparing for backend interviews with real project examples

Who it’s not for

Complete beginners who haven’t written C# yet
Anyone who wants slides and theory without code

FAQ

How do I cancel?

It’s a monthly Gumroad membership. Cancel anytime from your receipt’s “Manage membership” link or your Gumroad Library. You keep access until the end of the period you paid for.

Do I keep the files I download?

Yes, files you download are yours to keep.

Do I need API keys to run the projects?

Many projects include offline tests with fake models, so the test suites run without a key. Connecting a real AI provider needs your own key.

Do I get the older projects too?

Yes. You get immediate access to the whole library as soon as you join.

Will the $5 price last?

No. It’s a limited-time price and the regular price is $10/month.

Start building what production needs

~~$10~~ normally $5/month

$5/month for a very limited time, normally $10 — grab it now before it goes back to $10.

Join Tech Skill Builder