LLM

LLM

LLM - Applications

LLM - Applications - Chat UI

Certification

Study target

AI Agent & Agentic AI Expertise

Advanced proficiency in AI agent frameworks (LangChain, LangGraph, AutoGen, CrewAI, or similar).

Deep understanding of multi-agent systems, agent communication protocols, and coordination mechanisms.

Extensive experience with Large Language Models (LLMs) and their integration into agent workflows.

Expert-level prompt engineering and chain-of-thought reasoning implementation.

Proven experience with autonomous agent architectures (ReAct, Plan-and-Execute, Reflection patterns).

Knowledge of agent memory systems, tool integration, and environment interaction.

AI-assisted Development Workflow

  • Coding agent

Cheat sheet

Cheat sheet - Downloading models from Hugging Face

  • Ensure Git LFS is installed.

    git lfs install

  • Clone the model repo

    # e.g. MODEL_ID=Qwen/Qwen3-8B-GGUF
git clone https://huggingface.co/$MODEL_ID

Cheat sheet - Model reference

Resources

Framework for LLM use case evaluation

Type of customer needExampleML Implementation (Yes/No/Depends)Type of ML Implementation
Repetitive tasks where a customer needs the same output for the same inputAdd my email across various forms onlineNoCreating a rules-based system is more than sufficient to help you with your outputs
Repetitive tasks where a customer needs different outputs for the same inputThe customer is in “discovery mode” and expects a new experience when taking the same action:
- Generate a new artwork per click
- StumbleUpon style exploration		Yes- Image generation LLMs
- Recommendation algorithms (collaborative filtering)
Repetitive tasks where a customer needs the same/similar output for different inputs- Grading essays
- Generating themes from customer feedback		DependsIf simple: rules-based system works.
If complex combinations:
- Classifiers
- Topic modelling
Use LLMs for patternless or one-off cases
Repetitive tasks where a customer needs different outputs for different inputs- Answering customer support questions
- Search		YesToo many permutations for rules-based systems. Consider:
- LLMs with retrieval-augmented generation (RAG)
- Decision trees for products such as search
Non-repetitive tasks with different outputsReview of a hotel/restaurantYesPre-LLMs needed specialized models:
- Recurrent neural networks (RNNs)
- LSTMs
LLMs are a great fit for this type of scenario

LLMOps

LLMOps - API

API - Model Provider

API - API Gateway

Coding agent

  • Providers

  • Workflow

    • Create a feature branch
    • Create a plan / roadmap in markdown - either manually or chat to LLM to create and flesh it out.
    • Then work through the plan, making frequent commits in local feature branch. You can go step by step (my approach, lets me micro manage / write code where its quicker for me than the llm), or just YOLO and ask Sonnet 3.7 / Gemini 2.5 pro to try and one shot (probably not a good idea if you are actually a developer, but the option is there!)
    • Commit messages are unimportant, "WIP" is fine, its just the local branch, and enables easy rollbacks.
    • You can of course follow changes in the terminal, but using the git window > local changes is often helpful, if you are following the step by step approach.
    • When done, squash merge the local feature branch, writing a meaningful, semantic commit message

Concepts

Context length

The maximum number of tokens that an LLM can process in a single input sequence.

In simpler terms, context length acts as the model’s attention span determining how much information it can consider at once when generating responses.

Training

Training - Pretraining

  • Train the model on a large corpus of text (raw text).
  • Pretraining creates an initial pretrained LLM, called a base or foundation model.

Training - Fine-tuning

  • Train the model to predict the next word in the text on labeled data, aka fine-tuning.

  • 2 most popular fine-tuning categories

    • Instruction fine-tuning

      Labeled dataset consists of instruction and answer pairs, such as a query to translate a text accompanied by the correct translation.

    • Classification fine-tuning

      Labeled dataset consists of text and associated labels, such as emails with spam or not spam labels.

DevcontainerReading