Carsten Bansemir
Software Architect
12+ years SAP. 20+ years software architecture. Available now.
12+
Years SAP
20+
Years architecture
50%
Time saved (see here — to the SAP Accelerator Toolkit)
100%
Available now
Requirements analysis, solution and technical design, hands-on development and technical guidance within the team — including customer workshops and go-live support. Enablement included: your team applies AI tools and structured methods on its own afterwards.
12+ years SAP Fiori, ABAP, BTP and MII/MES — for clients including Schwarz IT/Lidl, Zeppelin and ARYZTA. Functional focus: production (PP, PM), logistics (WM, eWM, YL), service processes (CS). From custom ABAP to Clean Core, from shopfloor to boardroom.
In IT since 2002: 20+ years of software architecture across 4 industries — insurance, logistics, food production and construction. Twelve of those years outside the SAP world (Java EE, WebSphere): approaches beyond the standard SAP way. Diplom-Informatiker (AI) — FAU Erlangen-Nürnberg.
Custom rule-based accelerator toolkit with 170,000 lines of code and 250+ SAP validation rules. Claude Code as a daily tool. No LLM at runtime, no external calls — neither your code nor your business data leaves your environment. Measurable results, not just buzzwords.
Six specialized applications. Rule-based, deterministic, no AI wrapper.
It started with a practical question: how do I build AI applications that I actually use in day-to-day work — not demos, but tools that hold up in a real client project and can be tested under real conditions?
The second question followed: where is the competitive advantage in that? Not in access to a model — everyone has that. It is in what goes into it: twelve years of SAP project experience, cast into rules that otherwise exist only in individual consultants' heads. A tool that applies that experience on every request instead of retelling it every time.
And the third, which decides whether it can be used at a client at all: how do I use AI without creating permanent fixed costs? A tool that burns tokens on every call does not add up at scale — and in many landscapes it will not get approved in the first place. So the AI sits in the building, not in the running: developed with Claude Code, shipped as a deterministic application without a single model call at runtime. Built once, used without limit, at zero running cost.
Most AI tools for SAP are LLM wrappers: prompt in, answer out. That breaks down the moment traceability matters — the same input yields a different result on the second run, and nobody can explain why. For a review meant to support a release decision, that is useless.
Rule-based here means static SAP domain rules and score-based classification instead of probabilities. Same input, same result, reproducible at any time — and every finding can be traced back to the rule that raised it. With no external calls, neither code nor business data leaves the client's environment.
All six applications share the same architecture: a rule layer of versioned SAP domain rules, each with an ID, a severity and a rationale; an engine layer that classifies, scores and checks input against those rules; and a formatter layer that produces structured JSON and readable Markdown.
Wherever it can be substantiated, a rule carries its reference to the SAP documentation. What the tool claims can be looked up. The test suites hold that together — every domain rule is backed by several test cases. Python and FastAPI, each app runs standalone, database optional.
Score-based classification of requirement tickets by effort, risk, and feasibility.
What it does Takes unstructured ticket text in German or English and produces a ten-part technical assessment: request type (10 types, 28 subtypes), open clarifications, solution options from over 60 templates, impact mapping, effort estimate, risks, implementation plan and test view — plus a customer-facing version. Ten domain packs, including EWM, yard logistics, service, MII/MES and FI/CO.
Benefit Turns “the customer wants something with deliveries” into a defensible effort statement with its assumptions spelled out. The customer-facing version saves the second round of translation.
Turns a Joule use case and an existing on-premise landscape into a complete enablement blueprint: target operating model, IAS/IPS identity, cloud connector and destination design, Joule Studio skills, governance and AI unit estimation.
What it does Turns a Joule use case and an existing on-premise landscape into an enablement blueprint: target operating model including UX pattern, IAS/IPS identity, cloud connector and destination design, Joule Studio skills and agents, governance and AI unit estimation — plus starter scaffolds and the alternative track via SAP AI Core or the ABAP AI SDK.
Benefit Answers the question that comes up in every Joule pre-sales conversation: what exactly do we have to do to make this run in our landscape? As a verifiable blueprint rather than a slide deck — every load-bearing statement reconciled against SAP primary documentation.
Embedded readiness accelerator. Checks an S/4HANA Cloud Private Edition (RISE) landscape for embedded Joule readiness — gap analysis, activation runbook and fillable activation artifacts.
What it does Takes the described current state of a RISE landscape and checks it per Joule capability — navigational, informational, transactional, analytical — against a catalogue of documented prerequisites. Output: a gap list, an ordered activation runbook and prepared artifacts as JSON and Markdown.
Benefit Settles before the project what RISE customers otherwise discover during it: which prerequisite is still missing, and in what order to switch things on. The sibling to the on-premise accelerator — that one is about reaching Joule at all, this one about enabling embedded Joule.
Automated requirements analysis based on field definitions and change history.
What it does Takes a field request such as “add field X to app Y” and turns it into a technical solution draft: change-type classification (13 types), impact analysis across nine SAP layers from table through CDS, BOPF/RAP and OData up to the UI, plus risk assessment and test cases.
Benefit The question “what else does this touch?” is no longer answered from memory. Instead of writing a draft you review one — and immediately see the layer you would otherwise have missed.
Static analysis with 250+ SAP-specific validation rules for ABAP, CDS, and Fiori.
What it does Automated review for ABAP, CDS views, RAP behavior definitions, UI5 XML views and controllers. Over 250 rules across 21 modules: deprecation detection, manifest.json validation, clean-core checks such as DML on SAP tables or upgrade safety of BDEF extensions, and consistency checks spanning multiple artifacts.
Benefit Finds what the UI5 linter cannot see — especially the cross-artifact breaks between CDS, BDEF and UI. Every finding carries a rule ID, a severity and a concrete fix, so it can be argued with rather than merely asserted.
Skeleton generation for RAP Business Objects and Fiori Elements apps from functional specs.
What it does Generates complete development skeletons from a business goal description: CDS views, behavior definitions with determinations and validations, service definitions, Fiori Elements configuration across six floorplans, and UI5 TypeScript freestyle scaffolds.
Benefit The part of a RAP/Fiori project that is always the same takes seconds instead of days — consistently named and annotated. The time saved goes into business logic, not boilerplate.
Based on experience: approx. 30–50% faster development cycles with higher quality.