Complex Salesforce, taught simply.

I'm Mustafa Aksu, and I'm a Salesforce ISV Partner: I build Configra, my own AppExchange product that lets sales reps create deal-specific product configurations directly on a quote — without polluting the Revenue Cloud catalog. It came straight out of real Revenue Cloud work and is prepared for AppExchange Security Review; submitting it is the next step.

Day to day, most of my work sits in Revenue Cloud and Quote-to-Cash — my own RLM product and a complete quote-to-cash build: the product catalog and its bundles, Expression Sets that decide what each customer can buy, pricing, quotes and orders, tiered discount approvals, contracts, and the integrations that carry a signed order onward — to SAP S/4HANA, DocuSign, Stripe, and the German finance stack (lexoffice, DATEV, CAMT.053 bank statements). On top of that I build Agentforce: a live, deployed agent that answers grounded in Data 360 and takes real action (it identifies the customer, explains a consumption anomaly, and logs the support Case), custom Apex exposed to Claude as MCP tools over Salesforce Hosted MCP Servers, and an agent loop that built an entire Salesforce org — seven objects, triggers, tests — verified by its own test runs. One rule runs through all of it: the language model extracts intent, code sets the numbers. An agent is only as good as the org and the data underneath it.

And because an agent that works near a quote, a price or a customer record has to be trusted, I built the proof around it too — the whole Agent Development Lifecycle, the part almost nobody publishes. PreFlight scans a customer's real case history and scores whether the data can carry an agent at all — before a single build hour. Prüfstand attacks the live agent with a pre-registered German red-team corpus and grades it with a deterministic verifier the model never controls. Agent Blast Radius computes what an agent's code can actually reach and caught a GDPR-labelled field slipping past the running user's permissions — with a real org, not me, as the judge. Nabz watches the live agent's pulse and, when quality drops, correlates it with the org's change stream to name the why — with an evidence grade, never a guess. And Aktenlage turns an agent's operational traces into an EU AI Act Article 26 evidence pack. Trust is something you engineer, not something you claim.

But I didn't start in tech. I spent around twenty years in education, including six as a guidance counselor at an international college. That's not a detour from what I do now — it's the engine behind it. Teaching for two decades makes you good at one specific thing: taking something genuinely complex, structuring it clearly, and explaining it to someone who's never seen it before. This whole site is that skill pointed at Salesforce.

I care about code quality and user adoption in equal measure — clean architecture nobody uses is a failure, and so is a popular feature built on sand. Playing bağlama, ney, and piano taught me that a single character, condition, or line of code can change how a whole system behaves. I try to stay calm under pressure, listen before I build, and justify every decision I make.

I build in public and I give back to the community. I founded and moderate the Salesforce Revenue Cloud | Germany LinkedIn group, and you can see all eleven of my builds — TechnoStore (Quote-to-Cash on Revenue Cloud across ten external systems), Configra (my own RLM managed package), HanseWatt (the live agent — a fictional DACH energy retailer in a demo org), Hospital Org (an org built by an agent loop, with a router over three specialist agents), Agent Blast Radius (what an agent's code can really reach), Urla Shoes (AI document classification over MCP), VoltStream (a compliance engine with an agent that cannot invent law), and Prüfstand, PreFlight, Nabz and Aktenlage (the trust layer around the agents, closing the Agent Development Lifecycle) — on my Projects page, each linking to its full case study. Everything I write here comes from work I've actually shipped, not theory I read somewhere.