LLM Systems Engineering Portfolio

A collection of practical implementations demonstrating different LLM system architectures: Retrieval-Augmented Generation (RAG), Fine-Tuning (PEFT/LoRA), and Agentic Workflows.

Project Structure

Project	Type	Description
Chess RAG	Retrieval-Augmented Generation	A RAG system that answers questions about chess openings using a vector database (FAISS) and Semantic Search.
DialogSum LoRA	Fine-Tuning (QLoRA)	Parameter-Efficient Fine-Tuning of a generic LLM (TinyLlama) to perform dialogue summarization using the DialogSum dataset.
Weather Agent	Agentic Workflow	An autonomous agent capable of tool calling (function calling) to fetch real-time weather data and reason about it.

Tech Stack

• Language: Python 3.11
• Manager: uv
• Core Libraries: torch, transformers, peft, bitsandbytes, langchain, google-generativeai

Quick Start

This project uses uv for dependency management to ensure a reproducible environment.

1. Clone & Setup

# Clone the repository
git clone https://github.com/jpmalone0/llm-systems-engineering.git
cd llm-systems-engineering
# Install dependencies (creates .venv automatically)
uv sync

2. Configure Environment

Create a .env file in the root directory to store your API keys:

touch .env

Add your keys to .env:

GOOGLE_API_KEY=your_gemini_api_key_here
HF_TOKEN=your_huggingface_token_here

3. Run the Notebooks

Open the notebooks in VS Code or Jupyter Lab. Ensure you select the kernel named "Python (llm-systems-engineering)" or point to the .venv folder.

RAG: rag/chess_rag.ipynb

Fine-Tuning: fine-tuning-qlora/lora-dialogsum.ipynb

Agent: agentic-workflow/weather-agent.ipynb