The GenAI Lab

Anthropic's flagship model — excellent for coding, analysis, long-context reasoning, and agentic workflows. Claude Sonnet 4 is my primary work environment.

OpenAI's consumer AI — GPT-4o and o3 models with voice, image generation, browsing, code interpreter, and custom GPTs.

Google's multimodal AI — Gemini 2.5 Pro with 1M token context, Deep Research, and tight integration with Google Workspace.

Chinese open-weight frontier model — DeepSeek-R2 with exceptional reasoning and coding at a fraction of the cost of competitors.

xAI's model with real-time X (Twitter) data access, Grok 3 with deep reasoning mode, and image generation via Aurora.

Mistral AI's chat interface — fast, European, and multilingual. Mistral Large 2 and the open-source Mistral 7B family.

AI-native search engine with real-time web retrieval. Great for research, citations, and grounded answers with source links.

Open-source chat interface by Hugging Face — run Llama 3.3, Mistral, Qwen, and Command R+ for free with web search.

The most popular LLM application framework — chains, retrievers, memory, and 100+ integrations. Foundation for most production RAG pipelines.

Data framework for connecting private data to LLMs — indexing, querying, and 100+ data connectors. Best for RAG-heavy applications.

Open-source model hub, Transformers library, PEFT, datasets, and Spaces — the central ecosystem for open-weight AI development.

Type-safe agent framework from the Pydantic team — structured outputs, dependency injection, and model-agnostic agent layer with validation built in.

TypeScript-first SDK for building AI-powered web apps — streaming, tool use, React hooks, and support for all major LLM providers.

Graph-based stateful agent framework — nodes, edges, and checkpointing for complex multi-step reasoning, human-in-the-loop, and production workflows.

Role-based multi-agent collaboration — define crews of agents with tasks and goals. Adopted by 60%+ of Fortune 500 for enterprise agent deployments.

Microsoft Research's multi-agent conversation framework — agents debate, collaborate, and review each other's outputs. Best for research-grade workflows.

Production-grade replacement for Swarm — explicit agent handoffs, built-in tracing, and first-party tool integrations for OpenAI-native deployments.

Google's Agent Development Kit — hierarchical agent trees, native Vertex AI and Gemini integration, and A2A protocol for cross-framework agent communication.

Hugging Face's lightweight agent library — minimal code, code-writing agents, and model-agnostic tool use. Great for fast prototyping.

The developer standard for local LLMs — one command to pull and run Llama 3.3, Qwen3, Mistral, DeepSeek, and more with an OpenAI-compatible API.

Best GUI for local LLMs — visual model browser, download manager, and a local server that mimics OpenAI's API. Excellent on Apple Silicon.

100% offline desktop AI — clean ChatGPT-style UI, zero telemetry, and agentic Project workspaces with Browser MCP for fully private AI workflows.

Self-hosted ChatGPT-style interface wrapping Ollama or any OpenAI-compatible backend — multi-user, RAG, plugins, and conversation history.

High-throughput inference engine for production — PagedAttention for GPU memory efficiency, best-in-class tool calling, and OpenAI-compatible serving at scale.

Deep-dive survey posts by ex-OpenAI VP of Research — transformers, diffusion models, RL, and agent memory. Essential for conceptual depth.

Hands-on LLM experiments, tool use, and prompt injection research — one of the best builder-focused technical voices in the GenAI space.

Sebastian Raschka's newsletter — LLM architecture comparisons, training deep dives, and rigorous code-first explanations of cutting-edge research.

Real-world MLOps and production ML systems — AI engineering patterns, inference optimization, and hard-won lessons from building LLM products.

Applied ML, recsys, and LLM engineering from Amazon Principal Scientist — bridges academia and product with practical patterns for shipping AI systems.

Zero-to-hero neural network series, GPT from scratch, and state-of-LLM lectures — the single best deep learning resource on YouTube.

Long-form interviews with Sam Altman, Ilya Sutskever, Yann LeCun, and other AI leaders — philosophy, research, and the future of intelligence.

Visual deep learning — animated explanations of neural networks, attention, and transformers that make abstract math intuitive and memorable.

In-depth ML paper walkthroughs — transformers, reasoning models, and RL explained with whiteboard-level rigour for serious practitioners.

Fast, engaging breakdowns of the latest AI research papers — Károly Zsolnai-Fehér distills key ideas from complex work into digestible 5-minute videos.

Andrew Ng & Isa Fulford's free short course — iterative prompt development, summarisation, inference, transformation, and chatbot building with the OpenAI API.

End-to-end RAG course — chunking, vector databases, hybrid search, reranking, evaluation, and production deployment. ~5hrs/week for one month.

Build routers, tool-calling agents, and multi-document research assistants with LlamaIndex. Taught by Jerry Liu, LlamaIndex co-founder and CEO.

Learn document loading, splitting, embeddings, vector stores, retrieval, and Q&A chains using LangChain — a hands-on intro to RAG applications.

Andrew Ng's 5-course foundational program — neural networks, CNNs, sequence models, and structuring ML projects. The canonical entry point to deep learning.

Andrew Ng's non-technical GenAI course — how LLMs work, prompt engineering, real-world applications, and responsible AI use. Rated 4.8★ with 3300+ reviews.

Purpose-built search API for AI agents — relevance filtering, answer extraction, and structured results ready for LLM consumption. 1M+ downloads. Acquired by Nebius (2026).

Neural semantic search API for agents — find pages by meaning rather than keywords, with full content extraction. Strong for technical doc retrieval.

Independent search index — not a Google/Bing wrapper. Top scorer in 2026 agentic benchmarks (14.89). LLM Context API optimised for agent consumption.

Perplexity's API for grounded LLM answers — Sonar and Sonar Pro return LLM-synthesised responses with cited sources. Best for factual Q&A agents.

Google Search API wrapper — fast, cheap at scale ($0.30–$1/1k queries), JSON results, and a 2500-query free tier. Great for cost-sensitive high-volume agents.

Web scraping and crawling API for LLMs — turns any URL into clean markdown, handles JS-rendered pages, and pairs with search APIs for full-content agent pipelines.