Accepting new freelance contracts

Munich, Germany
Remote worldwide

MirwanAkaygün.

AI Automation Engineer · Systems, pipelines & API orchestration

Independent since 2019. I build and operate production AI automation: LLM-driven pipelines, API orchestration and the Linux infrastructure underneath — designed, shipped and kept running by one person.

LLM IntegrationPythonFastAPIRAG / pgvectorREST / OAuth2DockerLinuxTypeScript

Email me a brief

View selected work↓

Munich, DE

Remote / Hybrid

From 110 EUR/h

DE · EN

~/profile.json

{
  "name": "Mirwan Akaygün",
  "role": "AI Automation Engineer",
  "based": "Munich, Germany",
  "independent_since": 2019,
  "stack": [
    "LLM Integration", "Python",
    "FastAPI", "RAG / pgvector",
    "Docker", "Linux"
  ],
  "rate": "110 EUR/h",
  "available": true
}

Scroll

LLM IntegrationOpenAIAnthropicFunction CallingStructured OutputsPydanticPrompt EngineeringContext EngineeringLangGraphMCPRAGEmbeddingspgvectorPineconepromptfooLangfusePythonasyncioFastAPIREST APIshttpxOAuth 2.0WebhooksData PipelinesPostgreSQLSQLiteDockerLinuxCaddysystemdGitHub ActionsAWS BedrockMonitoringTypeScriptReactNext.jsGit

// approach

Fromasingleemailtosoftwarerunninginproduction.Architecture,code,tests,deploymentandthe24/7monitoring,ownedbyoneengineerendtoend.

// how

I don’t rely on a model doing what the prompt says. I engineer the conditions so it cannot do anything else: context scoped to the decision at hand, a deliberately bounded tool surface, schema-enforced output, response prefilling, and a validation gate every result has to clear before it moves downstream.

Scoping the context is not a limitation, it is the whole point. Everything you hand a model is one more thing it can get wrong, and one more thing you pay for on every single run.

A model with only one option doesn’t need discipline.

// 01

Stack & Expertise

Nine areas across AI integration, retrieval, evaluation, backend, data, infrastructure and delivery.

AI & LLM

LLM integration into production systems · OpenAI and Anthropic APIs called directly, no framework · function calling and structured outputs (Pydantic) · agentic tool chains

Prompt & Context

prompt engineering (few-shot, chain-of-thought) · context engineering · schema-enforced output · response prefilling · LangGraph · MCP (Model Context Protocol)

RAG & Retrieval

embeddings · pgvector · Pinecone · chunking strategies · PDF and Markdown parsing · retrieval pipelines · working understanding of fine-tuning

Evaluation & Observability

prompt evaluation (promptfoo) · LLM-as-a-judge · Langfuse tracing · cost and quality measurement per model call · automated regression checks

Backend & APIs

Python (OOP, asyncio) · FastAPI · REST API design · requests and httpx · OAuth 2.0 · webhooks and event-driven architectures · rate limiting, retry and idempotency

Data

PostgreSQL (incl. pgvector) · MySQL · SQLite · data modelling · deduplication · caching · reconciliation across large catalogues

Infrastructure

Linux · Docker · Caddy · systemd · cron and daemons · TLS automation · mail stack (Postfix, Dovecot, OpenDKIM) · fail2ban · monitoring

DevOps & Cloud

Git · GitHub Actions (CI/CD) · containerised builds and deploys · zero-downtime restarts · cloud fundamentals (AWS Bedrock, GCP)

Frontend & Method

Next.js · React · React Native (Expo) · TypeScript · Angular · measurement-driven optimisation · error handling and restart logic · documented handover

// 02

Selected Work

Production engagements covering backend, frontend, AI integration and mobile.

Production LLM Pipeline

Multi-stage Python pipeline with LLM-backed processing and an automated expected-versus-actual check that validates every run before it ships. Runs unattended with error handling and automatic restart. Hard cost ceilings per run, every model call logged and costed.

PythonLLM OrchestrationPipelinesQA Automation

$ python orchestrate.py

✓ OAuth refresh ok

✓ Rate limiter armed

Running batch...

✓ All sessions healthy

→ Scheduled next run

Multi-Account Trading Automation

Integration of multiple trading and supplier APIs over REST and OAuth2, with token lifecycle, rate limiting, retry and deduplication logic. Automated order and tracking reconciliation across a six-figure catalogue. Peak load of 36 transactions per day in unattended 24/7 operation, with structured logging and restart handling.

PythonREST APIsOAuth 2.0Data Pipelines

$ python etl.py

✓ Source ingested

✓ Schema validated

Transforming...

✓ Published

→ 0 validation errors

Self-Operated Production Infrastructure

Own Linux estate running every project above: Caddy with wildcard TLS and on-demand certificates, Postfix, Dovecot and OpenDKIM with SPF, DKIM and DMARC, fail2ban with custom filters, SSH hardening with key-only authentication, log rotation and monitoring. No managed hosting in between.

LinuxCaddyMail StackSecurity Hardening

$ pytest -x --parallel

tests/unit ........... ok

tests/integration .... ok

tests/e2e ............ ok

✓ All passed

→ Coverage report ready

Production Web Platforms

Several live platforms built and operated end to end: nexatalent.de for IT recruiting, bauenx.de for construction services with fixed-price offers, saunova.de with SEO structure and Search Console integration, and akay-immo.de with property presentation and enquiry flows. Further products run at meetyoursite.com and prowna.com. Plus a cross-platform mobile app in React Native, Expo and TypeScript.

Next.jsTypeScriptReact NativeSEO

$ python rag_query.py

✓ Embeddings loaded

Routing to provider...

✓ Structured output

✓ Validation passed

→ Response ready

// 03

How I Work

Async-first delivery model: scope by email, code via PR, no calls required.

Scope

You email the requirements. I review and ask clarifying questions within 24 hours.

Estimate

I send a written estimate with timeline, approach, and hourly breakdown. No call needed.

Build

Weekly deliverables via Git. You review PRs, I iterate. Structured async updates.

Ship

Production-ready code, documented, tested. Clean handoff with zero dependencies on me.

// 04

About

In software since 2016 (vocational training in application development at CREALOGIX, Puchheim, school grade 1.8). Continuously freelance since 2019. Backend, frontend and integration work for clients across DACH and the US.

Fully async workflow: you send requirements via email or ticket, I deliver production-ready code with clean git history. Registered self-employed in Germany, invoicing worldwide.

2016.

In software since

2019.

Independent since

DACH.

Markets

EN/DE.

Languages

// 05

Good fit if you need

You want an LLM integrated into a real business process, not a demo

You need an automation pipeline that runs unattended and reports what it costs

You need several APIs connected reliably, with token handling, retries and reconciliation

You want one person who builds it and also keeps it running in production

// 06

Common questions

The seven questions that usually come up before a brief lands.

Within 24 hours on business days. You send the scope in writing, I reply with clarifying questions, a written estimate and a proposed timeline. No discovery call required.

Hourly at 110 EUR/h is the default for ongoing work. Day rate 880 EUR for full-day blocks. Fixed price for clearly scoped deliverables. Invoices monthly with net 14 terms.

Yes. NDAs, MSAs and SOWs are routine. I invoice as a registered self-employed professional in Germany, with a German VAT ID for EU clients and reverse charge where applicable.

Remote-first. Occasional on-site days in Munich and the wider DACH region are possible by arrangement. Time zone overlap with EU and US East Coast works without friction.

Your repository, your pipelines. Weekly PRs with reviewable diffs, structured commit history, written architecture notes. Production-ready code, documented, tested. Zero dependency on me after handover.

From single 10-hour fixes to multi-month builds. No retainer minimum. If the brief is outside my stack, I say so up front instead of stretching.

German (native, C2) or English (professional working proficiency, C1). Documentation and code review in either, your preference.

// 07

Have a project in mind?
Let's build it.

Send me the scope via email. I respond within 24 hours with an estimate and approach. No calls required.

From 110 EUR/hDay rate 680 EURRemote, Munich, DACH

info@akayguen.dev Lebenslauf Resume