AI Extension Map

AI Builder Overview

Nobody stays a beginner forever. After the first deploy, AI features need retrieval evidence, evaluation criteria, repeatable procedures, and model optionality.

Recommended order

Start with a harness, add RAG, turn repeated work into skills, then compare local LLM options.

Run the 4-Day Course Start with Harness Recipe

Core Tracks

AI Harness

Lv.2

Repeatedly validate LLM features with golden sets, rubrics, and failure reports.

RAG Basics

Lv.3

Build document ingestion, chunking, embeddings, retrieval, and cited answers.

AI Skills

Lv.3

Freeze repeated work into reusable procedures and tool boundaries.

Local LLM

Lv.4

Compare Ollama and vLLM through OpenAI-compatible API patterns.

Expansion Keyword Radar

Product Surfaces

document QAsupport botticket triagecode review assistantcontent QAinternal search

Retrieval/Data

embeddingschunkingmetadata filtershybrid searchrerankingvector storeknowledge refresh

Agents/Workflow

tool callingMCPhuman-in-the-loopmemorybackground jobsapproval stepssandboxing

Evaluation/Quality

golden setprompt regressionsynthetic testsjudge modeltraceserror taxonomyrelease gate

Inference/Infra

OllamavLLMllama.cppGGUFquantizationGPU/VRAMmodel routingprompt cache

Security/Ops

PII redactionaudit logsaccess controldata retentionprompt injectionfallback modelcost guardrail

What should you build first?

You have many documents or internal knowledge

Start with RAG and citations, but keep a harness to measure retrieval quality.

You repeat the same operational work

Use skills and tool calling to lock inputs, outputs, and validation. Add MCP when external systems grow.

Cost, privacy, or offline use matters

Use Ollama for quick local tests, then compare vLLM when you need GPU-server inference.

Last updated: June 22, 2026 · Version: v0.0.1

Send Feedback

Opens a new issue page with your message.

Open GitHub Issue