AI 1003 · AI Engineering Core Skills

AI 1003

AI 1003 ·middot; Six modules · 13 tutorials · requires Python

Hands-on LLM engineering — from API calls and tokenization through open-source models, RAG pipelines, and QLoRA fine-tuning to production deployment. Build real applications at every stage: a website summariser, a multi-modal AI assistant, a meeting minutes generator, a code translation benchmark, a production RAG knowledge worker, and a deployed ensemble agent system.

Prerequisites: Python programming. Familiarity with APIs. AI 1001 (AI for Beginners) recommended. Google Colab access for GPU-dependent modules.

Code: AI 1003Level: Intermediate–AdvancedProvider: Universitas Scholarium

Module 1LLM Fundamentals2 tutorials

Drawing on the foundational research of Geoffrey Hinton

Transformer architecture · tokenization · context windows · API integration (OpenAI, Anthropic, Google) · running local models with Ollama · building a website summariser · chaining API calls · streaming output.

Open module →

Module 2Building with LLMs2 tutorials

Hosted by the Marvin Minsky Simulacrum

Gradio interfaces · multi-model integration · tool calling and function execution · SQLite database integration · multi-modal applications (DALL-E, TTS) · agentic workflows.

Open module →

Module 3Open-Source Models and HuggingFace2 tutorials

Drawing on the research of Yann LeCun in open-source AI

The HuggingFace ecosystem · Pipelines API · Google Colab with GPU · tokenizer internals · transformer architecture deep dive · quantization (4-bit, 8-bit) · building a meeting minutes generator from audio.

Open module →

Module 4Model Selection and Code Generation2 tutorials

Drawing on the evaluation methodology of Demis Hassabis

Benchmarks and leaderboards · the Chinchilla scaling law · model selection strategy · Python-to-C++ translation (230x speedup) · Python-to-Rust translation · frontier vs open-source comparison · technical metrics vs business outcomes.

Open module →

Module 5Retrieval Augmented Generation2 tutorials

Hosted by the Claude Shannon Simulacrum

Vector embeddings · chunking strategies · Chroma and FAISS · LangChain pipelines · evaluation (MRR, nDCG, LLM-as-judge) · advanced techniques (re-ranking, query expansion, GraphRAG) · from 0.73 to 0.91 MRR through systematic optimisation.

Open module →

Module 6Fine-Tuning and Deployment3 tutorials

Drawing on the training methodology of Yoshua Bengio

Dataset curation · baseline models (linear regression to XGBoost to neural networks) · frontier fine-tuning (OpenAI SFT) · QLoRA fine-tuning of LLaMA · Weights & Biases monitoring · serverless deployment with Modal · ensemble agents · the capstone: a deployed multi-model application.

Open module →