Generative AI & LLMs

How large language models work: tokens, context windows, attention, sampling parameters, and the sovereign distinction between open-weight and proprietary API-dependent models.

Craft effective prompts for sovereign LLM deployments: chain-of-thought, few-shot, system prompt design, prompt versioning, and structured output with JSON mode and Pydantic.

Sovereign open-weight model selection and deployment: Llama 4, Mistral, Qwen 3, Gemma 3, and Phi-4. Covers licensing analysis, quantisation formats (GGUF/AWQ), and model capability benchmarks.

Build sovereign multimodal AI pipelines: vision-language models (LLaVA, Qwen2-VL, Moondream), local image+text reasoning, audio transcription with Whisper, and zero-cloud multimodal stacks.

Integrate LLMs into sovereign applications via APIs and SDKs: OpenAI-compatible APIs (Ollama, vLLM), the Anthropic Python SDK, streaming responses, token counting, and API cost control.