Prompt Like a Pro: Tools, Techniques & Code for Perfecting LLM Prompts

Why Optimize Prompts?

• Get more accurate outputs
• Save on token cost
• Improve speed and reliability
• Enable dynamic prompting at scale

🔧 Libraries and Tools for Prompt Optimization

Below are powerful tools that can help you automate, experiment, and optimize prompt engineering — for research, production, or fun!

🧪 1. LangChain — Prompt Templates & Chains

LangChain is a full-stack framework for working with LLMs, and its PromptTemplate class is a core feature for structured prompting.

✅ Features:

• Templates with variables
• Few-shot examples
• Output parsers
• Evaluation chains

💡 Code Example:

from langchain.prompts import PromptTemplate

template = PromptTemplate.from_template(
    "Translate the following English to French: {text}"
)
prompt = template.format(text="I love AI.")
print(prompt)

Translate the following English to French: I love AI.

🔗 Docs: LangChain Prompt Templates

⚙️ 2. Guidance — Fine-Grained Prompt Programming

By Microsoft. It allows tight control over prompt execution, templating, and structured generation.

✅ Features:

• Mix programming and prompting
• Looping logic in prompts
• Structured output enforcement

💡 Code Example:

from guidance import gen, assistant,user,models
# load a model (could be Transformers, LlamaCpp, VertexAI, OpenAI...)
gpt = models.OpenAI("gpt-4",api_key="Your-API-Key")
with user():
    lm = gpt + "What is the capital of France?"

with assistant():
    lm += gen("capital")

🔗 GitHub: Guidance

📏 3. PromptLayer — Logging & Prompt Experimentation

PromptLayer lets you track, version, and compare prompts with OpenAI and other providers.

✅ Features:

• Monitor prompts & results
• Log metadata
• A/B testing prompts

💡 Code Integration:

import os
from promptlayer import PromptLayer

os.environ["OPENAI_API_KEY"] = "Your-API-Key"

promptlayer_client = PromptLayer(api_key="Your-API-Key")
OpenAI = promptlayer_client.openai.OpenAI
client = OpenAI()

prompt_text = "Explain gravity simply."

metadata = {
    "experiment": "gravity_experiment",
    "user": "aditya",
    "run_date": "2025-05-23"
}

response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": prompt_text}],
    pl_tags=["gravity", "test_run"],
    metadata=metadata,
    store=True
)

print(response.choices[0].message.content)

🔗 PromptLayer

🧪 4. LlamaIndex (fka GPT Index) — Prompt for RAG & Graphs

Best for combining data with LLMs (think RAG). Helps with prompt abstraction across indices and nodes.

✅ Features:

• Custom prompt templates for different node types
• Chaining prompts
• Works with LangChain to

💡 Prompt Template:

from llama_index.core.prompts import RichPromptTemplate

# Define the prompt template with placeholders for context and query
template = RichPromptTemplate(
    """We have provided context information below.
---------------------
{{ context_str }}
---------------------
Given this information, please answer the question: {{ query_str }}
"""
)

# Suppose you have your context and query like these:
context_str = (
    "The moon has no capital because it is not a country, "
    "but Earth's natural satellite."
)
query_str = "What is the capital of the moon?"

# Format the prompt as a plain string (good for completion APIs)
prompt_str = template.format(context_str=context_str, query_str=query_str)
print("Formatted Prompt String:\n", prompt_str)

# Format the prompt as a list of chat messages (for chat APIs)
messages = template.format_messages(context_str=context_str, query_str=query_str)

print("\nFormatted Messages:")
for message in messages:
    print(f"{message.role}: {message.content}")

Formatted Prompt String:
 We have provided context information below.
---------------------
The moon has no capital because it is not a country, but Earth's natural satellite.
---------------------
Given this information, please answer the question: What is the capital of the moon?

Formatted Messages:
user: We have provided context information below.
---------------------
The moon has no capital because it is not a country, but Earth's natural satellite.
---------------------
Given this information, please answer the question: What is the capital of the moon?

🔗 LlamaIndex Docs

🧰 5. OpenPrompt — Research-Grade Prompt Tuning

Academic-style prompt framework supporting manual, soft, and auto prompt tuning.

✅ Features:

• Plug-and-play with HuggingFace models
• Soft prompt tuning (gradient-based!)
• Few-shot experiments

🔗 GitHub: OpenPrompt

🧠 6. Promptify — Simplified Prompt Template & Evaluation Tool

Lightweight, easy-to-use Python tool for dynamic prompts.

💡 Code Example:

from promptify import Prompter,OpenAI, Pipeline

sentence     =  """The patient is a 93-year-old female with a medical       
                history of chronic right hip pain, osteoporosis,     
                hypertension, depression, and chronic atrial      
                fibrillation admitted for evaluation and management    
                of severe nausea and vomiting and urinary tract    
                infection"""

model        = OpenAI(openai_api_key) 
prompter     = Prompter('ner.jinja') # select a template or provide custom template
pipe         = Pipeline(prompter , model)

output = pipe.fit(text_input=sentence, domain="medical", labels=None)
print(output)

100%|██████████| 1/1 [00:07<00:00,  7.46s/it]
[{'text': "[{'T': 'Age', 'E': '93-year-old'},\n {'T': 'Gender', 'E': 'female'},\n {'T': 'Medical Condition', 'E': 'chronic right hip pain'},\n {'T': 'Medical Condition', 'E': 'osteoporosis'},\n {'T': 'Medical Condition', 'E': 'hypertension'},\n {'T': 'Medical Condition', 'E': 'depression'},\n {'T': 'Medical Condition', 'E': 'chronic atrial fibrillation'},\n {'T': 'Symptom', 'E': 'severe nausea and vomiting'},\n {'T': 'Disease', 'E': 'urinary tract infection'},\n {'branch': 'evaluation and management', 'group': 'admitted for'}]", 'usage': {'prompt_tokens': 205, 'completion_tokens': 156, 'total_tokens': 361, 'prompt_tokens_details': <OpenAIObject at 0x1e906e9d940> JSON: {
  "audio_tokens": 0,
  "cached_tokens": 0
}, 'completion_tokens_details': <OpenAIObject at 0x1e905b4dfd0> JSON: {
  "accepted_prediction_tokens": 0,
  "audio_tokens": 0,
  "reasoning_tokens": 0,
  "rejected_prediction_tokens": 0
}}, 'parsed': {'status': 'completed', 'object_type': <class 'list'>, 'data': {'completion': [{'T': 'Age', 'E': '93-year-old'}, {'T': 'Gender', 'E': 'female'}, {'T': 'Medical Condition', 'E': 'chronic right hip pain'}, {'T': 'Medical Condition', 'E': 'osteoporosis'}, {'T': 'Medical Condition', 'E': 'hypertension'}, {'T': 'Medical Condition', 'E': 'depression'}, {'T': 'Medical Condition', 'E': 'chronic atrial fibrillation'}, {'T': 'Symptom', 'E': 'severe nausea and vomiting'}, {'T': 'Disease', 'E': 'urinary tract infection'}, {'branch': 'evaluation and management', 'group': 'admitted for'}], 'suggestions': []}}}]

output[0]['parsed']['data']['completion']

[{'T': 'Age', 'E': '93-year-old'},
 {'T': 'Gender', 'E': 'female'},
 {'T': 'Medical Condition', 'E': 'chronic right hip pain'},
 {'T': 'Medical Condition', 'E': 'osteoporosis'},
 {'T': 'Medical Condition', 'E': 'hypertension'},
 {'T': 'Medical Condition', 'E': 'depression'},
 {'T': 'Medical Condition', 'E': 'chronic atrial fibrillation'},
 {'T': 'Symptom', 'E': 'severe nausea and vomiting'},
 {'T': 'Disease', 'E': 'urinary tract infection'},
 {'branch': 'evaluation and management', 'group': 'admitted for'}]

🔗 Promptify GitHub

🎯 Real-World Prompt Optimization Pipeline

Let’s bring this all together in a mini-pipeline

from langchain.prompts import PromptTemplate
import openai  # ✅ Correct import

# Template your prompt using LangChain
prompt = PromptTemplate.from_template("Summarize the following in 3 key points:\n\n{text}")

# Format your dynamic input
formatted_prompt = prompt.format(text="Artificial intelligence is the simulation of human intelligence in machines that are programmed to think and learn like humans. It enables machines to perform tasks such as problem-solving, decision-making, and language understanding.")

# Set your OpenAI API key (if not using environment variable)
openai.api_key = 'Your-API-Key'  # Replace with your actual key

# Call the OpenAI model
response = openai.ChatCompletion.create(
    model="gpt-4",
    messages=[
        {"role": "system", "content": "Be precise and neutral."},
        {"role": "user", "content": formatted_prompt}
    ]
)

# Evaluate or log the output
print("\n✨ Summary:\n")
print(response["choices"][0]["message"]["content"])

✨ Summary:

1. Artificial intelligence (AI) imitates human intelligence in machines by programming them to think and learn like humans.
2. AI allows machines to execute tasks such as problem-solving and decision-making.
3. These tasks also include language understanding, signifying the machine's ability to interpret and respond to linguistic inputs.

Conclusion

The humble prompt has evolved from a simple instruction to a programmable, tunable, testable interface to intelligence.

With tools like LangChain, Guidance, and PromptLayer, you can:

• Test different prompts
• Automate dynamic flows
• Evaluate and compare
• Integrate with RAG, agents, and apps