AI & LLM integration patterns
Call LLM APIs, structure prompts, parse responses, and ship AI features safely.
Cache LLM Completions by Hashing the Prompt in Python
A simple in-memory cache that stores LLM completions keyed by a SHA-256 hash of the prompt to avoid recomputing identical requests.
import hashlib
import json
class PromptCache:
def __init__(self):
self.cache = {}
def _hash_prompt(self, prompt: str) -> str:
return hashlib.sha256(prompt.encode("utf-8")).hexdigest()
def get(self, prompt: str) -> str | None:
key = self._hash_prompt(prompt)
return self.ca…Cosine Similarity to Retrieve Top K Chunks in Python
Compute cosine similarity between a query vector and a list of chunk vectors, then return the indices and scores of the top k most similar chunks.
import numpy as np
from numpy.linalg import norm
def cosine_similarity(vec1, vec2):
return np.dot(vec1, vec2) / (norm(vec1) * norm(vec2))
def retrieve_top_k(query_vec, chunk_vectors, k=3):
similarities = [cosine_similarity(query_vec, vec) for vec in chunk_vectors]
top_indices = sorted(range(len(similarit…How to Build a Zero-Shot Classification Prompt in Python
Creates a prompt for zero-shot text classification by pairing input text with candidate labels and a hypothesis template.
from typing import Dict, List
def build_zero_shot_prompt(
text: str,
candidate_labels: List[str],
hypothesis_template: str = "This is about {}.",
) -> Dict[str, List[str]]:
"""Build a prompt ready for zero-shot classification."""
return {
"sequences": text,
"candidate_labels": can…How to Compute a Mock BLEU Score with n-gram Overlap in Python
Evaluate text similarity with a simplified BLEU score using word-level n-gram precision and a brevity penalty.
from collections import Counter
def bleu_score(reference, candidate, n=2):
"""
Compute a simplified BLEU score with n-gram precision and brevity penalty.
Mock demo using word-level n-grams.
"""
ref_tokens = reference.lower().split()
cand_tokens = candidate.lower().split()
# Compute n-…How to Detect Prompt Injection in Python
Implements a regex-based heuristic in Python to flag common prompt injection attempts before sending input to an LLM.
import re
def contains_prompt_injection(user_input: str) -> bool:
# Directives to ignore previous instructions or act as system
ignore_patterns = [
r"\bignore\s+(all\s+)?previous\s+instructions\b",
r"\bdisregard\s+(all\s+)?previous\s+instructions\b",
r"\bdon'?t\s+follow\s+(any\s+)?inst…How to Filter Blocked Words in Python
Scans input text against a moderation blocklist, returning blocked terms and their counts.
MODERATION_BLOCKLIST = {"spam", "scam", "fraud", "phishing", "malware", "abuse"}
def scan_text(text: str) -> dict:
normalized = text.lower()
words = normalized.replace(".", " ").replace(",", " ").replace("!", " ").replace("?", " ").split()
found_terms = []
for word in words:
if word in MO…How to Filter Toxic Keywords in Python
Filter toxic keywords from text by replacing each occurrence with asterisks, useful as a basic guardrail for LLM inputs.
TOXIC_KEYWORDS = ["insult", "threat", "hate", "violence", "spam"]
def guardrails_filter(text: str, keywords: list[str] | None = None) -> str:
"""Filter out toxic keywords from the given text.
Args:
text: The input text to filter.
keywords: Optional keyword list. Defaults to TOXIC_KEYWORDS.
…How to Log Prompts and Completions as JSONL Audit Files in Python
Read a JSONL file of LLM prompt–completion pairs, compute totals and averages, then write an audit summary with timestamps.
import json
from pathlib import Path
from datetime import datetime
def audit_jsonl(filepath):
logs = []
with open(filepath, encoding="utf-8") as f:
for line in f:
line = line.strip()
if not line:
continue
entry = json.loads(line)
logs.ap…How to Parse JSON from LLM Model Output Fence in Python
Extract and parse a JSON object from a language model's output that may be wrapped in triple-backtick fences with an optional language tag.
import json
import re
def parse_json_from_fence(text):
"""
Extract JSON object from a model output that may be wrapped in
triple-backtick fences with optional language tag.
"""
# Match content insideHow to Validate JSON Output Against a Dict Schema in Python
Validate JSON-like data against a simple dict schema with type checking and descriptive error messages using only the Python standard library.
from typing import Dict, Any, List, Union
def validate_json(data: Any, schema: Dict[str, str]) -> List[str]:
"""
Validate JSON-like data against a simple dict schema.
Schema format: {field_name: expected_type} where type is one of:
'str', 'int', 'float', 'bool', 'list', 'dict', 'any'
Returns list …How to Validate LLM Output in Python
A beginner-friendly DataValidator class that checks required fields and type constraints on LLM-generated or user JSON data.
import json
from typing import Any, Dict, List, Optional
class DataValidator:
"""Simple helper for validating LLM-generated or user data."""
def __init__(self, required_fields: List[str], schema: Optional[Dict[str, str]] = None):
self.required_fields = required_fields
self.schema = schema or…How to cache embeddings with a Python dict to avoid recomputation
Caches embeddings computed from text in a dictionary keyed by SHA-256 hash, returning cached results for repeated calls.
import hashlib
import time
class EmbeddingCache:
def __init__(self):
self.cache = {}
def _hash_text(self, text):
return hashlib.sha256(text.encode()).hexdigest()
def get_embedding(self, text, compute_func):
key = self._hash_text(text)
if key not in self.cache:
…How to compute ROUGE recall in Python
Compute ROUGE recall by counting token overlap between a reference and candidate summary with pure Python.
def rouge_recall(reference, candidate):
ref_tokens = reference.lower().split()
cand_tokens = candidate.lower().split()
ref_counts = {}
for token in ref_tokens:
ref_counts[token] = ref_counts.get(token, 0) + 1
cand_counts = {}
for token in cand_tokens:
cand_counts[token] = cand…How to compute exact match metric in Python
Computes the exact match (EM) metric for LLM outputs by normalizing text and comparing predictions against references.
def compute_exact_match(predictions, references):
def normalize(text):
import re
text = text.lower().strip()
text = re.sub(r'\b(a|an|the)\b', ' ', text)
text = re.sub(r'[^a-z0-9\s]', '', text)
text = ' '.join(text.split())
return text
matches = sum(1 for pred, r…How to parallel map embeddings with a thread pool in Python
Run embedding computations in parallel using ThreadPoolExecutor, collect results into a dict keyed by the original item.
import threading
from concurrent.futures import ThreadPoolExecutor
import time
def compute_embedding(item: int) -> tuple[int, int]:
time.sleep(0.05) # Simulate embedding work
return item, item * 10
def parallel_map_embed(items, max_workers=3):
results = {}
with ThreadPoolExecutor(max_workers=max_w…How to parse JSON in Python: A Beginner's Guide with Code Examples
This guide shows you how to parse JSON data in Python step by step, with practical code examples and expected outputs.
import json
from typing import Any, Dict, List, Optional
class DataHelper:
"""Beginner-friendly helper for common AI/LLM data tasks."""
def __init__(self, data: Optional[Dict[str, Any]] = None):
self.data = data or {}
def to_prompt(self, template: str) -> str:
"""Format a prompt…JSON Mode Prompt Schema Output in Python
Extract a user object to JSON with explicit schema keys, ready for LLM JSON-mode prompts.
import json
from typing import Any, Dict
def extract_user_as_json(user: Dict[str, Any]) -> str:
"""Extract a user object and return it as JSON using explicit schema keys."""
schema_fields = ("id", "name", "email", "is_active")
user_subset = {key: user[key] for key in schema_fields if key in user}
ret…Browse by section
Each section groups closely related Python snippets.
AI & LLM integration patterns — Python code examples
What you will find here
This page collects ai & llm integration patterns snippets — short, copy-ready Python you can paste into our free online IDE and run without installing anything. Each sample includes a plain-English explanation and the full source code.
Samples vs tutorials and challenges
Samples are quick reference — one concept per page. For step-by-step teaching, use our Python tutorials. To test yourself, try quizzes or coding challenges. Clean up style with the Python formatter.