The Understanding Open AI Workspaces course is for developers with intermediate machine learning experience and Python skills who are new to Generative AI and want to learn how to build, customize, optimize, and deploy open source large language models.

This course provides learners with the skills to set up, configure, and manage environments for open generative AI development. Beginning with local installations, learners practice running large language models on their own machines using Ollama, exploring performance optimization techniques for consumer hardware, and integrating external applications through APIs. The course then introduces Docker and Docker Compose, guiding learners through containerized environments that ensure reproducibility, persistence, and scalability. Learners build multi-container architectures to separate models and services while managing GPU passthrough and memory optimization. Finally, the course covers Google Colab for cloud-based GPU access, where learners configure free resources, manage storage through Google Drive, and monitor performance within session constraints. By the end, learners will have set up both local and cloud environments, documented their processes, and gained the ability to choose the right workspace for different AI workloads.

The Preparing Text for AI Models course is designed for developers, engineers, and technical product builders who are new to Generative AI but already possess intermediate machine learning knowledge, basic Python proficiency, and familiarity with development environments such as VS Code, and who want to engineer, customize, and deploy open generative AI solutions while avoiding vendor lock-in.

The course equips learners with practical skills in dataset sourcing, preprocessing, and formatting for training large language models. Starting with the discovery of text datasets from repositories like Hugging Face, Kaggle, and Common Crawl, learners evaluate quality, relevance, and licensing considerations. The course then covers preprocessing pipelines, including text cleaning, normalization, deduplication, and tokenization strategies, ensuring efficiency and compatibility with model training. Learners also design annotation schemas, apply semi-automated labeling techniques, and build validation workflows to maintain quality. The final module guides learners in constructing structured datasets for instruction tuning, fine-tuning, and benchmarking, supported by best practices in train-test splits and stratification. By the end of the course, learners will have created production-ready text datasets suitable for generative AI applications.

The Fine-Tuning Text Models with PEFT course introduces learners to parameter-efficient fine-tuning methods that enable large language model adaptation on limited hardware. Learners start with foundational concepts of PEFT and Low-Rank Adaptation (LoRA), understanding their advantages over full fine-tuning in terms of memory, cost, and flexibility.

The course then dives into implementing QLoRA, combining quantization with LoRA for high-performance fine-tuning on consumer GPUs. Learners practice setting up training environments, preparing datasets, optimizing hyperparameters, and managing checkpoints. The final module emphasizes evaluation, using metrics such as perplexity, BLEU, ROUGE, and BERTScore to measure improvements. By the end, learners will have implemented a fine-tuning pipeline and produced a domain-adapted LLM with performance documentation.

The Fine-Tuning Image Models with Diffusion course gives learners hands-on experience adapting generative image models for custom styles and applications. The course begins with the foundations of diffusion models, explaining forward and reverse diffusion processes and exploring the key components of Stable Diffusion architectures, including U-Net, VAE, and text encoders. Learners then apply Low-Rank Adaptation (LoRA) techniques to train efficiently on consumer hardware, comparing performance and trade-offs with full fine-tuning. In the second module, learners implement DreamBooth, a methodology for training on limited datasets to personalize models with custom concepts and artistic styles. They practice dataset preparation, hyperparameter tuning, and checkpoint management while preserving model generalization. The final module introduces ComfyUI, where learners design and execute node-based workflows that integrate fine-tuned models with advanced extensions like ControlNet.

By the end of the course, learners will have produced a custom fine-tuned diffusion model, integrated it into ComfyUI pipelines, and optimized it for production-quality image generation.

The Model Evaluation and Benchmarking course equips learners with the skills to assess and compare the performance of both text and image generative models. Starting with text evaluation, learners apply standard metrics such as perplexity, BLEU, ROUGE, and BERTScore, while also designing human evaluation protocols and task-specific methods for applications like summarization or translation.

The course then explores image evaluation using technical metrics, including FID, CLIP similarity, and SSIM, alongside human perception-based assessment techniques and artifact detection systems. In the final module, learners design comprehensive benchmarking frameworks with reproducible testing environments, version control, and visualization dashboards for continuous monitoring. By the end, learners will be able to implement automated, domain-specific evaluation systems and deliver detailed performance reports that ensure generative models meet rigorous quality standards.

The Optimizing Models for Production course prepares learners to make generative AI models more efficient, scalable, and cost-effective for real-world deployment. Learners begin with quantization, applying INT8 and INT4 precision reduction using tools like bitsandbytes while balancing accuracy and efficiency. Next, they explore inference optimization strategies, including batching, KV-cache management, and token-level computation scheduling to reduce latency in interactive applications.

The course also covers memory footprint reduction and adaptive batch sizing for dynamic workloads. In the final module, learners apply practical hardware optimization techniques such as GPU memory tuning, mixed precision inference, and profiling tools like nvidia-smi and PyTorch Profiler to identify bottlenecks. By the end, learners will be able to deliver optimized models across diverse hardware environments, supported by performance benchmarks and reproducible deployment pipelines.

The Building RAG Systems with Open Models course provides learners with the skills to design and implement retrieval-augmented generation (RAG) applications for real-world use cases. Learners start by exploring the fundamentals of RAG architecture, breaking down key components such as retrievers, rankers, generators, and orchestration layers, while learning design patterns for tasks like question answering, summarization, and knowledge synthesis.

They then dive into embeddings and vector databases, comparing FAISS, ChromaDB, Milvus, and Pinecone, and applying indexing and chunking strategies to improve retrieval efficiency and semantic relevance. The final module brings all elements together to build production-ready RAG pipelines using LangChain and open LLMs, incorporating advanced retrieval methods, hallucination mitigation, and evaluation frameworks for accuracy and reliability. By the end, learners will have built a functional RAG application with configurable components, optimized for performance and equipped with robust evaluation metrics.

The API Development and Model Serving course teaches learners how to deploy and expose generative AI models through robust and scalable APIs. Beginning with FastAPI, learners design and implement REST endpoints for model inference, focusing on schema design, authentication, rate limiting, and error handling.

The course then introduces the Model Context Protocol (MCP), comparing it with traditional API approaches and demonstrating how function calling and tool integration can extend model capabilities. In the final module, learners address scaling and performance, applying containerization with Docker, asynchronous request handling, load balancing, and monitoring techniques. Practical exercises also cover tunneling and remote access using ngrok for rapid prototyping. By the end, learners will have built a production-ready API with clear documentation and the ability to support both REST and MCP-inspired integration patterns, equipping them with the tools to serve generative AI applications efficiently and reliably.

The AI Agent Development Fundamentals course introduces learners to the core design patterns and practical skills required to build autonomous AI agents. Learners begin by studying the architectural foundations of agent systems, including perception, reasoning, and action loops, as well as the differences between reactive, deliberative, and hybrid agent types.

The course then focuses on building simple reactive agents, where learners apply structured prompting, decision-making frameworks, and natural language understanding to implement predictable and testable behaviors. In the final module, learners extend their agents with tool-use and memory management capabilities, using function-calling patterns, conversation history maintenance, and context window optimization. Practical exercises emphasize building agents with resilience through error handling and recovery strategies. By the end of the course, learners will have created functional agents capable of integrating tools, maintaining memory, and performing autonomous tasks.

The Deploying Open Models course teaches learners how to package, host, and maintain generative AI models in real-world production environments. The course begins with Docker containerization, where learners design optimized Dockerfiles, apply dependency management techniques, and implement security practices such as isolation and access control.

Next, learners explore cloud deployment strategies, comparing options across AWS, GCP, Azure, and specialized providers, while also evaluating cost, performance, and compliance considerations. They will also gain hands-on experience with rapid prototyping on Hugging Face Spaces and learn about serverless architectures for efficiency. In the final module, the focus shifts to monitoring and maintenance, where learners implement logging systems, performance dashboards, alerting frameworks, and version control practices to ensure reliable long-term operations. By the end of the course, learners will have deployed an open model with comprehensive monitoring, security, and update management in place.

The Ethics and Safety in Open AI course equips learners with the frameworks and tools needed to ensure responsible use of generative AI models. The course begins with bias detection and mitigation, where learners identify harmful patterns in datasets and outputs, apply quantitative evaluation techniques, and implement mitigation strategies.

Next, learners design and test safety guardrails, including input validation, output filtering, content moderation, and red-teaming practices to strengthen AI systems against misuse. The final module covers content provenance, licensing, and compliance, where learners apply watermarking techniques, implement provenance standards such as C2PA, and evaluate datasets and models for licensing adherence. Regulatory frameworks like GDPR and CCPA are also introduced. Through hands-on exercises, learners will build safety layers, implement provenance metadata, and prepare compliance-ready audit documentation. By the end, learners will be able to design open AI applications that prioritize safety, fairness, and accountability.