Generative AI Projects and Lessons

This repository is a public portfolio of Generative AI projects, teaching examples, and exploratory notebooks by Douglas Daly.

The work here reflects two related goals:

1. Build practical AI systems that connect LLMs, retrieval, agents, structured outputs, evaluation, observability, and human review into useful workflows.
2. Teach modern AI concepts clearly through small, reproducible lessons that explain how the patterns work, where they help, and where they break.

My focus is practical AI architecture: systems that are understandable, traceable, testable, and useful beyond the first demo.

Start here

• Project index: docs/project_index.md
• Lesson index: docs/lesson_index.md
• Featured project: projects/resume_builder/
• Featured lesson: lessons/a2a/

Some examples require API keys, external datasets, larger model downloads, GPU acceleration, or additional setup. Check the relevant project or lesson README before running a full workflow.

Repository structure

projects/   End-to-end AI projects and portfolio systems
lessons/    Concept-focused teaching examples and walkthroughs
notebooks/  Jupyter notebooks for exploration, instruction, and prototypes
src/        Reusable Python package code shared across projects and lessons
docs/       Project indexes, lesson indexes, notes, and quickstarts
assets/     Small supporting assets used by examples

Projects

The projects/ folder contains larger AI systems organized around practical workflows or product-style artifacts. These are meant to show architecture judgment, implementation tradeoffs, and end-to-end design.

Examples include:

• AI-Assisted Resume Generation System Evidence-driven targeted resume generation using canonical JSON, market-derived role signals, structured LLM workflows, human review checkpoints, versioned artifacts, and DOCX/PDF rendering controls.

• GraphRAG and Coverage Assistant Traceable retrieval and structured decision support over connected evidence, entities, clauses, and reasoning paths.

• Archival Restore AI-assisted restoration and organization workflows for messy real-world media and metadata tasks.

A folder belongs in projects/ when the primary goal is to build or simulate a useful AI-enabled workflow.

Lessons

The lessons/ folder contains smaller, concept-first examples designed for teaching and experimentation. These examples are intentionally narrower than the projects and are meant to be easy to run, inspect, and adapt.

Lesson topics include:

• Agent-to-agent coordination
• Model Context Protocol concepts
• Tool calling and structured outputs
• Retrieval-augmented generation
• Multi-agent orchestration
• Multimodal AI workflows
• Evaluation and regression detection
• Diffusion, CLIP, VAEs, GANs, and representation learning

A folder belongs in lessons/ when the primary goal is to explain a concept, framework, protocol, or implementation pattern.

Featured project: AI-Assisted Resume Generation System

The resume builder is an evidence-driven pipeline for creating targeted resume variants from a structured source resume.

It separates:

• Source evidence
• Canonical resume data
• Market-derived role signals
• Evidence-to-signal mapping
• Content selection
• Human review
• DOCX/PDF rendering

The goal is not to automate away judgment. The goal is to make the resume generation process more traceable, reviewable, and repeatable.

Featured lesson: Agent-to-Agent Coordination

The A2A lessons introduce agent-to-agent coordination patterns through small, inspectable examples.

The goal is to show how specialized agents can expose capabilities, how a coordinating agent can route work, and how the system can log decisions and outputs for review.

Local setup

git clone https://github.com/dougdaly/Generative-AI.git
cd Generative-AI
conda env create -f environment_min.yml
conda activate genai

Alternatively, use a Python virtual environment and requirements.txt:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Open the notebooks in Jupyter or VS Code. Use the project and lesson indexes to choose a starting point.

Repository notes

• Keep examples reproducible where practical.
• Include sample outputs when they help reviewers understand the workflow.
• Do not commit large model checkpoints, private data, secrets, or licensed datasets.
• Use projects/ for end-to-end systems and lessons/ for concept-focused teaching examples.
• This repository is curated as a public portfolio and teaching resource.

License

MIT. See the LICENSE file.