Oliver Muellerklein

Oliver Muellerklein¶

I build agentic systems end-to-end: LLM-driven extraction pipelines, semantic search, custom MCP tooling, and the full-stack applications that wrap them. Background in environmental science, geospatial data, and HPC.

See Waggle → · TRACE on GitHub → · Email

Selected work¶

Waggle: agentic CRM

A full-stack agentic CRM I built for a beekeeping wholesale domain. Custom chatbots, an automated ordering pipeline that converts inbound email into typed orders, and a semantic-search RAG layer over operational data. Live app and architecture diagrams.

Open source

TRACE: provenance for AI-assisted workflows

I built TRACE, an open-source MCP-based Python package for tracking who made what decision and why in AI-assisted workflows. If you've thought hard about agent observability, audit trails, or provenance in AI-assisted dev, I want collaborators. Star, open an issue, or send a PR.

★ Star on GitHub → Read the spec ↗

Skills¶

AI / agentic systems

LLM tool use, prompt engineering, eval design
Custom MCP server design (Python)
LangGraph + ReAct agents
RAG, semantic search, embeddings (pgvector)
Decision provenance + audit trails (TRACE)

Backend / data engineering

Python: FastAPI, Pydantic, async pipelines
JavaScript / TypeScript, Clojure / ClojureScript
Postgres + pgvector, SQL, Alembic migrations
FSM design, idempotent ETL, schema-first ingest
Pyright, type-driven Python; Pytest with real-data fixtures

Cloud, HPC + containers

AWS (EC2, SageMaker), Azure, GCP
HPC: Slurm, batch jobs, distributed compute
Containers: Docker, Kubernetes
Apptainer (renamed from Singularity, 2021)
Reproducible env management across cloud + on-prem

Geospatial / mapping

PostGIS for spatial SQL
Google Earth Engine: low-level API + scripts
GDAL / OGR, raster + vector workflows
Interactive web maps + geospatial models in Python, JavaScript, Clojure / ClojureScript
Satellite imagery (Sentinel, Landsat), time-series rasters, CRS handling

Data science / R / applied ML

R + RStudio: extensively, ecological + environmental modelling
Python: pandas, scikit-learn
Time-series and spatial statistics
Reproducible analysis pipelines

Engineering practice

E2E testing against real data over mocks
Fail-loud over silent error swallowing
Schema-first design, Pydantic across boundaries
Open-source MCP servers and tooling
TRACE-style decision provenance for AI work