14-day roadmap across closed-source LLM APIs, open-source models & fine-tuning, and agentic AI frameworks. Your launchpad into AI engineering.
Build a character AI chatbot: first API call, response anatomy deep-dive, and persona engineering techniques to give your bot a unique personality.
Build an AI calorie tracker using the Vision API. Master prompt engineering patterns and image-to-nutrition analysis with active recall exercises.
Build an AI tutor UI with Gradio: wire up system prompts, add an explanation-level slider, and stream responses — from bare function to polished web app.
Master the full multi-LLM landscape: open vs closed models, nine key benchmark numbers, token math, calling multiple provider SDKs, and risk-proofing your stack.
13-part deep dive into the open-source LLM stack: Hugging Face ecosystem, GPU/VRAM architecture, quantization (GGUF/GPTQ), KV cache, tokenization, generation pipeline, RAG architecture, and chat templates — from API consumer to architecture decision-maker.
When to reach for a reasoning model vs a standard LLM — covers o1/o3 and open-source thinking models, selection frameworks, and the benchmarks that actually predict task performance.
Build a full RAG pipeline with LangChain: document loaders, text splitters, embedding models, vector stores, retrieval chains — wired together with LCEL for production-ready retrieval.
End-to-end RAG pipeline notes: chunking strategies, embedding models, vector database indexing, retrieval & reranking, and evaluation techniques for reducing hallucinations.
Build an AI resume assistant with Pydantic structured output and multi-provider LLMs (OpenAI/Gemini): parse resumes, generate tailored cover letters, and extract structured career insights from unstructured text.
Hands-on fine-tuning with LoRA and PEFT on open-source models: adapter layers, QLoRA memory optimisation, custom dataset prep, and the full Hugging Face Trainer loop.
Build multi-agent teams with AutoGen: define agent roles, wire up communication patterns, and orchestrate complex task flows where agents collaborate rather than just respond.
Build stateful agentic workflows with LangGraph: model control flow as a graph, add cycles and conditional routing, and persist agent state across multi-step tasks.
Automate the full data science workflow with CrewAI: define specialist agents (analyst, engineer, modeller), wire tasks with shared context, and let the crew drive predictive analytics end-to-end.
Build powerful agentic AI workflows visually with n8n: connect LLMs to real-world triggers, chain multi-step automations, and integrate with hundreds of services — no boilerplate needed.
Build an advanced AI tutor that uses MCP to pull context from external tools and the OpenAI Agents SDK to run a robust agentic loop — tool calling, context injection, and multi-step reasoning.
When prompting isn't enough: LoRA, QLoRA, and PEFT explained — plus instruction tuning, dataset quality criteria, and a decision framework for choosing fine-tuning vs RAG vs prompt engineering.
Knowing if your LLM system is actually good — benchmarks, LLM-as-judge, hallucination detection, and guardrails — then keeping it fast with vLLM, KV-cache, and a proper observability stack.
14 battle-tested blueprints covering the full prompt engineering landscape — LLM mechanics, prompting patterns (zero-shot to ReAct), transformer attention, and sampler control.
Architect-focused recall notes for prompt engineering: system prompts, context windows, instruction structure, and prompting techniques — designed for active retention.
MCP from first principles to enterprise scale: server/client/host model, tools vs resources vs prompts, architecture tradeoffs, and real-world patterns — distilled for architect-level recall.