Samuel Iso

Understand how large language models work — tokens, context windows, and inference.

• Explain how transformers generate text
• Understand tokenisation and context limits
• Compare GPT-4, Claude, Gemini, Llama capabilities

Build production apps with LLM APIs — function calling, structured output, streaming.

• Call LLM APIs with function/tool use
• Parse structured JSON from LLM output
• Handle streaming responses
• Implement retry/fallback logic

Fine-tune open-source LLMs with LoRA/QLoRA for custom tasks and domain adaptation.

• Prepare fine-tuning datasets
• Run LoRA/QLoRA training with Unsloth or HF Trainer
• Evaluate and merge adapters
• Push models to Hugging Face Hub

Understand the ReAct loop, agent memory, and how tool-using agents work.

• Explain the ReAct pattern
• Build a simple tool-calling agent
• Implement short-term memory with conversation history

Understand and operate vector databases — indexing, metadata filtering, ANN.

• Set up Pinecone, Weaviate, or pgvector
• Implement metadata filtering
• Benchmark ANN recall vs latency

Build a basic RAG pipeline — chunk, embed, retrieve, and generate.

• Chunk documents with LangChain or LlamaIndex
• Generate embeddings and store in a vector DB
• Build a Q&A app over your own data

Build AI-powered automation pipelines with n8n, Make, or custom orchestrators.

• Design end-to-end autonomous pipelines
• Connect LLMs to external services via automation tools
• Monitor and recover from failures

Generate images with Midjourney, DALL-E 3, and Stable Diffusion.

• Generate photorealistic and stylised images with prompts
• Use negative prompts and style parameters
• Understand model differences (Midjourney vs SD vs DALL-E)

Use AI tools to scale content production, personalise campaigns, and automate marketing.

• Generate and repurpose content at scale with AI
• Personalise email sequences with AI segmentation
• Automate social media scheduling with AI tools

Build and optimise email lists, sequences, and automation flows.

• Build a 5-email welcome sequence
• Achieve >40% open rate with subject line testing
• Set up Mailchimp/Klaviyo automation flows

Validate ideas, find problem-solution fit, and ship an MVP fast.

• Run 5 customer discovery interviews
• Define a minimal testable hypothesis
• Ship a no-code or coded MVP in two weeks

Use Copilot, Cursor, and Claude Code to write and refactor code faster.

• Use inline completions and chat in Cursor/VS Code
• Write effective prompts for code generation
• Refactor legacy code with AI assistance