Skip to content
A.C.M.P. GmbH
[AI Automation]

Processes that run themselves. When you build them right.

From audit to workflow automation to AI agents — we build the systems that replace manual work without anyone clicking through a GUI.

Done
Generating · 64%
Done
Processing
Done
Generating · 28%
Pipeline · live6 / 24 in queue
[01]Automation

Workflow & Document Automation

Invoice extraction, contract analysis, lead routing, internal approval flows. Instead of SaaS lock-in with per-user pricing, we build the logic directly into your stack — visible, testable, version-controlled.

What we deliver

  • Document parsing (invoices, contracts, IDs)
  • Multi-step workflows with retries and audit log
  • Integrations with existing systems (ERP, CRM, storage)
  • Dashboards for running processes

Tools we use

  • Inngest or Temporal for workflow orchestration
  • GPT-4o + Claude Vision for document AI
  • Postgres as the single source of truth
  • TypeScript end-to-end — no YAML, no GUI
Workflow · running1.4k events / hr
[02]Consulting

Process Audit & Redesign

Before we automate, we look closely. Which steps are redundant, which are invisible, which actually need a human? Redesign first, then tooling — otherwise you just automate the mess.

What we deliver

  • Interviews with process owners and end users
  • End-to-end mapping of existing processes
  • Redesign proposal with ROI estimate
  • Implementation roadmap with quick wins first

Tools we use

  • Structured interviews and workshops
  • Process mining on existing logs
  • Mermaid diagrams in Git instead of Visio in SharePoint
  • Prototypes instead of PowerPoint slides
Funnel · last 30 days24.6k → 980
[03]Agents

AI Agents & Decision Support

Agents that classify, summarize, route, or make small decisions on their own. RAG over your knowledge, tool-use for real actions — built on an architecture you understand and can debug.

What we deliver

  • RAG systems over internal documents
  • Sales and support assistants with tool-use
  • Classification and routing agents
  • Evaluation setups that catch hallucinations

Tools we use

  • OpenAI + Anthropic for reasoning
  • pgvector or Pinecone for retrieval
  • LangGraph or Vercel AI SDK
  • Our own evals — no blind trust
[04]Our stack

Best-in-class. Nothing else.

  • OpenAI

    GPT-4o for reasoning, code generation, and document AI. Our default for structured outputs.

  • Anthropic

    Claude for long contexts, precise classification, and nuanced tool-use.

  • Inngest

    Durable functions for long-running workflows with retries, steps, and side effects.

  • Temporal

    When Inngest isn't enough — state machines at enterprise scale, self-hostable.

  • Postgres

    Source of truth for data and workflow state. pgvector for embedding search.

  • Pinecone

    Managed vector search when Postgres doesn't cut it or latency matters.

  • Playwright

    Browser automation for legacy systems without an API. Headless, deterministic, versioned.

  • Browserbase

    Hosted browsers in the cloud — for scalable browser tasks and crawls.

  • Vercel AI SDK

    Streaming, tool-calls, multi-provider — directly in the React stack, no vendor lock-in.

[05]Use case

Example: Invoice Extraction Pipeline

Industrial · DACH · 2025

Manufacturing SMB

We automated the inbound invoice flow: PDFs in, structured data into the ERP. Accounting types nothing manually anymore — and sees every step in the audit log.

Invoices per month
0
Manual typing
0%
Processing time
0%

Ready for process automation?

30 minutes, no strings. We take apart one of your processes and show where AI would move the needle right away.

Response within 24 hours