LLM Experiments

This repository showcases practical implementations of various LLM techniques, from simple text generation to more complex agentic systems with self-execution capabilities and persistent memory. Each project is designed to demonstrate specific architectural patterns in LLM application development.

All projects are built in Python with the use of LangGraph and a local Ollama LLM. See the README in each project for

Setup

Unless stated otherwise in the project's README, the setup for all projects is:

1. Install Ollama from https://ollama.com
2. Run ollama pull llama3.2
3. Install Poetry from https://python-poetry.org/docs/#installing-with-pipx
4. Run poetry install

To execute the project, simply run poetry run start

Projects

1. Choose Your Own Adventure

Path: choose-your-own-adventure/

An interactive storytelling experience powered by LLMs that generates dynamic narratives based on user choices.

Key Features:

• Dynamic story generation with context preservation
• Adaptive narrative paths based on user input
• State management for consistent story progression

Techniques Demonstrated:

• Simple multi-step agentic workflows

---

2. Data Visualizer

Path: data-visualizer/

An intelligent data analysis tool that automatically generates Python code to analyze datasets and produces comprehensive HTML reports.

Key Features:

• Automatic file format detection and parsing
• LLM-generated Python code for custom analysis
• Self-execution of generated code with safety measures
• HTML report generation with visualizations

Techniques Demonstrated:

• Code generation and validation
• Sandboxed code execution
• Error handling and retry logic
• Multi-step agentic workflows

---

3. Mental Health RAG

Path: mental-health-rag/

A Retrieval-Augmented Generation (RAG) system that answers mental health questions using a curated knowledge base of mental health resources.

Key Features:

• Vector database integration for semantic search
• PDF document parsing and chunking
• Context-aware response generation

Techniques Demonstrated:

• Document embedding and indexing
• Vector similarity search (done under-the-hood by Chroma)
• RAG

Potential Future Projects

1. Note-Taking Assistant - Demonstrates long-term memory management
2. TBD - Demonstrates Agent-to-Agent protocol using Strands Agents
3. TBD - Demonstrates MCP using the official modelcontextprotocol SDK

---

For questions or collaboration opportunities, please reach out via LinkedIn.