Skip to content
HenkelHiedl
HenkelHiedl project work

HenkelHiedl

Machine Learning Engineer · Jul 2025 – Present · Berlin

henkelhiedl.com

HenkelHiedl is a Berlin-based design and strategy agency, founded in Kreuzberg in 2003, working with brands and institutions on digital communication, identity, and product.

I architected an internal LLM chatbot with MCP servers, integrating internal documents, a structured knowledge base, a skills layer, and a prompt library — giving staff a single interface to query institutional knowledge.

I also built a large-scale LLM-powered data pipeline to mine over 600,000 dental records for structured clinical data, surfacing patterns and insights spanning 20 years of patient history. The pipeline handles noisy, semi-structured clinical notes and extracts computable fields at scale.

Chatbot Architecture

4 integration layers behind a single chat interface. Staff query institutional knowledge, trigger agent skills, and access curated prompts.

MCP SERVERS
INGEST → CHUNK → EMBED → INDEX → RETRIEVE → AUGMENT → GENERATE → VALIDATE → INGEST → CHUNK → EMBED → INDEX → RETRIEVE → AUGMENT → GENERATE → VALIDATE →
LAYER 01RAG

Internal Docs

Company documents indexed for retrieval

LAYER 02VECTOR DB

Knowledge Base

Structured institutional knowledge

LAYER 03TOOLS

Skills Layer

Task-specific agent capabilities

LAYER 04TEMPLATES

Prompt Library

Curated prompt templates

DATA PIPELINE

600K+

dental records processed

20yr

patient history span

48

structured fields extracted