AI Engineering Cheat Sheet

AI Engineering Cheat Sheet - BelajarKoding | BelajarKoding

#Getting Started

Instalasi library utama yang dipake buat AI engineering dengan Python.

#Install Libraries

Cara install library AI yang paling sering dipake.

bash

# OpenAI Python SDK
pip install openai
 
# Anthropic SDK (Claude)
pip install anthropic

python

import anthropic
 
client = anthropic.Anthropic()
 
message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1000,
    system="Kamu adalah senior code reviewer yang memberi feedback singkat.",
    messages=

python

messages = [
    {"role": "user", "content": """
Klasifikasikan sentiment review berikut:
 
Review: 'Mantap, kualitas oke banget!' => positif
Review: 'Lama banget pengirimannya' => negatif
Review: 'Sesuai deskripsi, oke lah' => netral
 
Review: 'Rusak, kecewa berat' =>
"""}
]

python

system_prompt = """
Sebelum menjawab, pikirkan langkah demi langkah.
1. Identifikasi informasi yang diketahui
2. Identifikasi yang ditanyakan
3. Hitung langkah demi langkah
4. Berikan jawaban akhir dengan format: JAWABAN: [angka]
"""

python

# Code reviewer
"Kamu adalah senior developer. Review kode dengan fokus pada: bug, performance, readability. Berikan saran dengan contoh kode perbaikan."
 
# Customer support
"Kamu adalah customer support untuk [produk]. Jawab dengan ramah, singkat, dan akurat. Jika tidak tahu, arahkan ke human agent."
 
# Data extraction
"Ekstrak informasi dari teks. Kembalikan HANYA dalam format JSON dengan field: nama, tanggal, jumlah, mata_uang. Jangan tambahkan teks lain."
 
# Tutor
"Kamu adalah tutor programming. Jangan kasih jawaban langsung. Beri petunjuk dan biarkan murid berpikir."

python

from openai import OpenAI
 
client = OpenAI()
 
# Single text
response = client.embeddings.create(
    model="text-embedding-3-small",
    input="Belajar AI engineering untuk developer"
)
vector =

python

from sentence_transformers import SentenceTransformer
 
model = SentenceTransformer('all-MiniLM-L6-v2')  # 384 dimensi, cepat
 
embeddings = model.encode([
    "Laptop gaming murah",
    "Notebook untuk bermain game",
    "Resep nasi goreng spesial"
])
 
# Cek similarity

python

import numpy as np
 
def cosine_sim(a, b):
    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))
 
sim = cosine_sim(vector_a, vector_b)
# Range 0-1 (atau -1 sampai 1). Makin deket ke 1, makin mirip.

python

from langchain.text_splitter import RecursiveCharacterTextSplitter
 
splitter = RecursiveCharacterTextSplitter(
    chunk_size=500,       # Maksimal karakter per chunk
    chunk_overlap=50,     # Overlap antar chunk biar konteks nggak putus
    separators=["\n\n"

python

from langchain_community.document_loaders import TextLoader
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain_community.vectorstores import Chroma
from langchain.chains import

text

+------------+-------------+----------+------------+-----------+
| Database   | Type        | Language | Best For   | Open Src? |
+------------+-------------+----------+------------+-----------+
| Chroma     | Embedded    | Python   | Prototype  | Ya        |
| Pinecone   | Managed SaaS| Any      | Production | Tidak     |
| Weaviate   | Self-host   | Go/Python| Hybrid src | Ya        |
| Qdrant     | Self-host   | Rust API | High speed | Ya        |
| Milvus     | Distributed | Go/C++   | Scale      | Ya        |
| pgvector   | PG Extension| SQL      | Sudah pakai| Ya        |
| FAISS      | Library     | Python   | In-memory  | Ya        |
+------------+-------------+----------+------------+-----------+

python

from transformers import pipeline
 
# Sentiment analysis
classifier = pipeline("sentiment-analysis")
classifier("Filmnya bagus banget!")
# [{'label': 'POSITIVE', 'score': 0.999}]

python

from transformers import AutoTokenizer, AutoModelForCausalLM
 
model_name = "meta-llama/Llama-3.2-1B"
 
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)
 
# Generate text
inputs = tokenizer("Belajar AI itu", return_tensors=

python

from peft import LoraConfig, get_peft_model, TaskType
from transformers import AutoModelForCausalLM, AutoTokenizer
 
model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-3.2-1B",
    load_in_4bit=True,  # 4-bit quantization (QLoRA)

text

+------------------+-----------+----------------+----------+--------+
| Model            | Provider  | Context Window | API Cost | Open?  |
+------------------+-----------+----------------+----------+--------+
| GPT-4o           | OpenAI    | 128K tokens    | $$$      | Tidak  |
| GPT-4o mini      | OpenAI    | 128K tokens    | $        | Tidak  |
| Claude Sonnet 4  | Anthropic | 200K tokens    | $$       | Tidak  |
| Claude Haiku 3.5 | Anthropic | 200K tokens    | $        | Tidak  |
| Gemini 2.0 Flash | Google    | 1M tokens      | $        | Tidak  |
| Gemini 2.5 Pro   | Google    | 1M tokens      | $$       | Tidak  |

bash

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
 
# Download dan jalanin model
ollama run llama3.2          # Model 1B-3B
ollama run mistral            # Model 7B
ollama run qwen2.5            # Model Qwen
 
# Via API (Ollama menyediakan endpoint OpenAI-compatible)

python

# Pakai Ollama dari Python (OpenAI-compatible)
from openai import OpenAI
 
client = OpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama"  # Bebas, nggak dicek
)
 
response = client.chat.completions.create(
    model="llama3.2"

python

import tiktoken
 
enc = tiktoken.encoding_for_model("gpt-4o")
 
text = "Ini adalah contoh teks untuk dihitung tokennya."
tokens = enc.encode(text)
print(f"Jumlah token: {len(tokens)}")
 
# Estimasi biaya

python

from pydantic import BaseModel, ValidationError
 
class ProductInfo(BaseModel):
    name: str
    price: float
    stock: int
 
def extract_product(llm_output):
    """Parse dan validasi output LLM ke schema."""
    try:

text

Token          Unit pemrosesan LLM (~4 karakter / 3/4 kata)
Context Window Maksimal token yang bisa diproses sekaligus
Temperature    Kontrol kreativitas (0=konsisten, 1=kreatif)
Embedding      Representasi numerik teks yang nangkap makna
Vector DB      Database khusus simpan dan cari embeddings
RAG            Sistem ambil data eksternal, inject ke prompt
Fine-Tuning    Latih ulang model dengan data spesifik
LoRA           Fine-tuning efisien, update sedikit parameter
QLoRA          LoRA + quantization 4-bit, butuh GPU lebih kecil

import json
from openai import OpenAI
 
client = OpenAI()
 
# Definisikan tools
tools = [
    {
        "type": "function",
        "function": {
            "name": "get_product_info",
            "description": "Dapatkan info produk berdasarkan ID",
            "parameters": {
                "type": "object",
                "properties": {
                    "product_id": {

Baca Cheat Sheet Lengkap

#Getting Started

#Install Libraries

#Setup API Keys

#OpenAI API

#Basic Chat Completion

#Streaming Response

#Structured Output (JSON Mode)

#Vision API

#Anthropic API (Claude)

#Prompt Engineering Patterns

#Zero-Shot

#Few-Shot

#Chain-of-Thought

#System Prompt Patterns

#Embeddings

#OpenAI Embeddings

#Open Source Embeddings (Sentence Transformers)

#Cosine Similarity Manual

#RAG Pipeline

#Document Chunking

#RAG dengan ChromaDB

#RAG dengan LangChain

#Vector Databases Comparison

#Pinecone Setup

#pgvector (PostgreSQL Extension)

#Function Calling & Tools

#OpenAI Function Calling

#Hugging Face Transformers

#Pipeline API (Paling Gampang)

#Load Model Manual

#Vercel AI SDK (TypeScript)

#Install dan Setup

#Generate Text

#Streaming dengan React Hook

#Route Handler (Next.js App Router)

#Structured Output (Zod)

#Fine-Tuning dengan LoRA/QLoRA

#LoRA Config

#Training Loop dengan TRL

#LLM Model Quick Reference

#Model Serving

#Ollama (Local LLM)

#vLLM (High-Throughput Serving)

#Common Patterns

#Rate Limiting & Retry

#Token Counting

#Response Validation

#Conversation Memory

#Glossary Cepat