llm-agentic-workflow-sandbox

My personal learning sandbox for building LLM-based agentic workflows in pharmacometrics. Each numbered folder is a self-contained example; and future examples will build on what earlier ones established.

Built with LangGraph for orchestration, LangChain for provider-agnostic model access, and Pydantic for structured outputs and state validation.

This is a personal learning project, not production code. Expect rough edges and the occasional misunderstanding. My reasoning is documented in the accompanying blog posts.

Examples

• 01_hello_world/ — Two linear nodes (extractor → reporter) extracting PK parameters from a short paper. The foundation everything else will build on. Blog post: Building a Human-in-the-Loop LLM Agent Workflow for Pharmacometrics — Part 1.

More to follow soon: human-in-the-loop interrupts, conditional routing, NONMEM integration, ...

Setup

git clone https://github.com/marianklose/llm-agentic-workflow-sandbox.git
cd llm-agentic-workflow-sandbox

python -m venv .venv
source .venv\Scripts\activate          # Other: .venv/bin/activate

pip install -r requirements.txt

Create a .env file at the repo root with your API key(s):

ANTHROPIC_API_KEY=sk-ant-...
# OPENAI_API_KEY=sk-...             

Running an example

python 01_hello_world/main.py