Welcome to JojoJubah

What is Prompt Engineering?

Prompt engineering is the craft of structuring inputs to get reliable outputs from AI models. It's part instruction design, part debugging. Good prompts are clear, specific, and include examples.

Follow-up prompting (why it works)

LLMs refine answers when you add constraints or feedback. Ask the model to "improve this," "explain your reasoning briefly," or "compare options in a table." Iteration > one-shot perfection.

Few-shot examples (when to use)

When your task has a pattern, show 1—3 short input→output examples. Keep style consistent and label clearly. This anchors the model and reduces randomness.

Context windows & chunking

Long inputs get truncated. Prioritize: brief instructions → constraints → examples → source snippets. If content is large, chunk it and summarize progressively.

What is Context Engineering?

Context engineering is structuring the right info around your prompt so the model has what it needs (and nothing it doesn't). Think: goal → constraints → key facts → examples → attachments.

Core moves

• Be explicit about the outcome and format.
• Include only the relevant facts (links/files beat giant pasted walls).
• Ask the model to ask you clarifying questions first if anything is ambiguous.
• For long sources, chunk + summarize rather than dumping everything.

What is an AI Agent?

An agent is an LLM that can plan steps and use tools (like a browser, code runner, or files) to reach a goal. Think: "goal → plan → act → observe → iterate."

Tools, memory, planning

Tools extend capabilities (search, code, APIs). Short-term memory tracks the session. Long-term memory can store facts or preferences. Planners break big tasks into small executable steps.

Risks & guardrails

Agents can loop, hallucinate, or act out of scope. Use sandboxing, confirmation prompts for risky actions, strict tool schemas, and time/step limits.

What does "open-source LLM" mean?

Weights + code are available under a license you can run locally. Great for privacy, customization, and cost control—trade-off: may need more setup/tuning.

Popular local tools

Ollama (easy model pulls), LM Studio (GUI), Text-Gen WebUI (power user). Use GGUF quantized models for lower RAM.

Running on modest hardware

Start with 7B—8B models, use 4—8-bit quantization, prefer CPU-friendly backends, and keep context short. Cache prompts and chunk data for speed.