方寸 Portal

AI Engineer Portal

Your personal operating system for career transition.

Private mode
All Learning Paths
Learning Path

LLM App Foundations

Learn the product and systems primitives behind useful LLM applications.

0 / 12 lessons complete0%
16h total intermediate
Continue where you left off

Prompt, context, tools, and memory

Map the core components of an LLM application before choosing frameworks.

Start lesson
1Up next

Prompt, context, tools, and memory

Map the core components of an LLM application before choosing frameworks.

45 min
2

Request lifecycle of an LLM feature

Trace a user request from UI input to final response and logging.

45 min
3

Guardrails and structured outputs

Constrain outputs with schemas, system prompts, and approval boundaries.

50 min
4

Cost, latency, and failure budgets

Make tradeoffs visible instead of treating providers like magical black boxes.

45 min
5

Portfolio slice: ship one narrow assistant well

Design a focused feature instead of a vague chatbot.

45 min
6

Provider SDK integration patterns: OpenAI and Anthropic

Wire up production-grade clients for both major providers, handle auth, streaming, and error normalization, and build a provider-agnostic wrapper that survives API changes.

55 min
7

Prompt templating and multi-turn conversation management

Build reusable prompt templates with variable injection, manage multi-turn conversation state with proper history trimming, and implement the summarize-on-overflow pattern for long sessions.

55 min
8

Structured outputs, JSON mode, and response validation

Get reliably structured data out of LLMs using tool use, JSON mode, and Pydantic validation; implement defense-in-depth parsing for production features; handle model fallback chains for resilience.

60 min
9

LangChain fundamentals: chains, prompts, and parsers

Learn when LangChain adds value, how LCEL composes pipelines, and how to produce structured output with Pydantic parsers.

45 min
10

Provider abstraction patterns

Compare LangChain's model interface against direct SDK calls, understand the callback system, and decide when abstraction pays off.

40 min
11

Fine-tuning vs RAG: decision framework

A systematic decision framework for choosing between fine-tuning and RAG, covering dataset curation, LoRA/QLoRA, hybrid approaches, and the specific criteria that make each technique the right tool for the job.

40 min
12

Multi-modal AI for application engineers

Practical patterns for vision, audio, and document processing in production AI applications: sending images to vision models, speech-to-text and TTS integration, PDF and OCR pipelines, and building multi-modal chains.

40 min