21.07.2026 aktualisiert
100 % verfügbar
Senior AI Engineer — Agentic & Multi-Agent Systems
Eschede, Deutschland
Nur Remote
M.Sc. Artificial Intelligence and Distributed SystemsÜber mich
I take AI projects from notebook to production: agentic workflows, RAG, and the reliability layer underneath (eval gates, cost controls, observability). Solo, end-to-end. 18 months, 3 enterprise teams: 99.7% uptime, retrieval 72→91%. EU citizen, Frankfurt, remote. Freelance, available now.
Skills
RAGMLflowBedrockMLOpsVertex AILangChainApache AirflowAmazon Web ServicesMicrosoft AzureGoogle BigQueryPythonPostgreSQLNatural Language ProcessingRedisTensorFlow
Multi-Agent Systems and LLM Orchestration
Production multi-agent systems in Python with LangGraph and LangChain, DSPy prompt optimization, and Semantic Kernel. State-machine orchestration with checkpoint/resume, Pydantic-validated guardrails, and human-in-the-loop review for reliable long-running pipelines.
Agentic AI on AWS Bedrock
Bedrock AgentCore (Gateway, Runtime, Observability) with Model Context Protocol (MCP) tools, microVM per-tenant isolation, and AWS Lambda for deterministic execution. LLM routing across Claude, Llama, and Amazon Nova inside the customer's cloud.
RAG and Retrieval
Retrieval-augmented generation over Vertex AI Search, Azure AI Search, and Pinecone. Citation-forced grounding so every answer cites a source paragraph and hallucinations fail loudly.
Machine Learning and NLP
PyTorch transformer fine-tuning (BERT, DistilBERT), scikit-learn evaluation, and Pandas/NumPy pipelines. NLP classification systems and golden-dataset regression wired into CI/CD.
LLM Evaluation and Observability
Golden-dataset eval gates, Langfuse and Bedrock AgentCore Observability traces, and MLflow tracking. Per-tenant cost, latency, and drift visible in production; regressions blocked at merge.
Cloud Platforms
AWS, Azure, and GCP: Azure AI Foundry Agent Service, Vertex AI, GKE, and Cloud Run. Multi-cloud deployments keeping models and data inside the customer's sovereign environment.
DevOps and Infrastructure
Docker, Kubernetes, Terraform infrastructure-as-code, GitLab CI/CD, and Azure Container Registry image promotion. Reproducible staging-to-production stacks with signed builds.
Backend and Data
Python, FastAPI, and Pydantic. SQL and BigQuery, Redis for agent state and checkpointing, and Airflow for batch orchestration with automatic retries.
Production multi-agent systems in Python with LangGraph and LangChain, DSPy prompt optimization, and Semantic Kernel. State-machine orchestration with checkpoint/resume, Pydantic-validated guardrails, and human-in-the-loop review for reliable long-running pipelines.
Agentic AI on AWS Bedrock
Bedrock AgentCore (Gateway, Runtime, Observability) with Model Context Protocol (MCP) tools, microVM per-tenant isolation, and AWS Lambda for deterministic execution. LLM routing across Claude, Llama, and Amazon Nova inside the customer's cloud.
RAG and Retrieval
Retrieval-augmented generation over Vertex AI Search, Azure AI Search, and Pinecone. Citation-forced grounding so every answer cites a source paragraph and hallucinations fail loudly.
Machine Learning and NLP
PyTorch transformer fine-tuning (BERT, DistilBERT), scikit-learn evaluation, and Pandas/NumPy pipelines. NLP classification systems and golden-dataset regression wired into CI/CD.
LLM Evaluation and Observability
Golden-dataset eval gates, Langfuse and Bedrock AgentCore Observability traces, and MLflow tracking. Per-tenant cost, latency, and drift visible in production; regressions blocked at merge.
Cloud Platforms
AWS, Azure, and GCP: Azure AI Foundry Agent Service, Vertex AI, GKE, and Cloud Run. Multi-cloud deployments keeping models and data inside the customer's sovereign environment.
DevOps and Infrastructure
Docker, Kubernetes, Terraform infrastructure-as-code, GitLab CI/CD, and Azure Container Registry image promotion. Reproducible staging-to-production stacks with signed builds.
Backend and Data
Python, FastAPI, and Pydantic. SQL and BigQuery, Redis for agent state and checkpointing, and Airflow for batch orchestration with automatic retries.
Erfahrung
Berufserfahrungmehr als 10 Jahre
Freelancer seit2019
Sprachen
EnglischMutterspracheSpanischMuttersprache
Projekthistorie
Production multi-agent system on AWS that reads compliance documents and flags policy violations, so the team only reviews what the agents escalate. LangGraph and LangChain orchestration over Bedrock (Claude, Llama, Nova), Bedrock AgentCore with MCP tool calls and per-tenant microVM isolation, RAG grounded in governance data with Langfuse tracing, and a golden-dataset regression suite gating every GitLab release.
Multi-agent document-classification system on Azure across SAP, SharePoint, and HR systems, keeping every record inside the agency's sovereign cloud. Azure AI Foundry Agent Service with Semantic Kernel, a fine-tuned PyTorch transformer via Azure Machine Learning, and Azure OpenAI, Claude, and Llama behind a versioned Pydantic prompt layer, with RAG over Azure AI Search and a full SQL audit trail.
Agentic freight-quoting system on Google Cloud that returns binding quotes inside the customer's session window. LangChain on Vertex AI (PaLM 2, Gemini, Claude), RAG over the carrier and rate-sheet corpus via Vertex AI Search, DSPy prompt optimization, Docker microservices on GKE and Cloud Run, and BigQuery joins, with an LLM regression suite blocking any accuracy drop at merge.