DEV Community: rishisingh007

From Prompt Perfection to Iteration: The New Way to Use AI as a Developer

rishisingh007 — Tue, 19 Aug 2025 10:46:40 +0000

For years, people believed that writing a perfect prompt was the key to getting value out of AI tools like ChatGPT or Claude. But today’s reality is very different, with everyday advances in AI capabilities.

Modern AI models are much more capable, and we get to know this once we start using them. Even basic prompts often return production-quality results. Let us use two different approaches to reach to desired results. First we will start with a basic prompt and then refine it with next prompt, and record the result. In second approach we will provide full requirement details in the first prompt itself and take the results. Then compare the results of two approaches.

Let us start with our first approach.

First Approach

GPT Prompt: Write a function to check if a number is prime.

The result code in Python:

def is_prime(n: int) -> bool:
    """Check if a number is prime."""
    if n <= 1:
        return False
    if n <= 3:
        return True
    if n % 2 == 0 or n % 3 == 0:
        return False

    # Check divisibility up to sqrt(n)
    i = 5
    while i * i <= n:
        if n % i == 0 or n % (i + 2) == 0:
            return False
        i += 6
    return True

Look at the code is is already optimized code without ever mentioning in the prompt. Now let us ask GPT to give code for very large numbers in the next prompt to refine our result.

GPT Prompt: Make it efficient for very large numbers and suggest if I should use probabilistic methods.

The result code below:

import random

def is_probable_prime(n: int, k: int = 20) -> bool:
    """Miller-Rabin primality test with k iterations."""
    if n < 2:
        return False
    # Small primes
    for p in [2, 3, 5, 7, 11, 13, 17, 19, 23, 29]:
        if n % p == 0:
            return n == p

    # Write n-1 as d*2^s
    s, d = 0, n - 1
    while d % 2 == 0:
        s += 1
        d //= 2

    # Witness loop
    for _ in range(k):
        a = random.randrange(2, n - 1)
        x = pow(a, d, n)
        if x == 1 or x == n - 1:
            continue
        for _ in range(s - 1):
            x = pow(x, 2, n)
            if x == n - 1:
                break
        else:
            return False
    return True

Second approach

Now let us combine above two prompts into one, thus making it a complex prompt first time.

GPT Prompt: Write a function to check if a number is prime. Make it efficient for very large numbers and suggest if I should use probabilistic methods.

The result code below:

import random

def is_prime_miller_rabin(n: int, k: int = 10) -> bool:
    """Miller-Rabin primality test. 
       k = number of rounds (higher = more accuracy)."""
    if n < 2:
        return False
    if n in (2, 3):
        return True
    if n % 2 == 0:
        return False

    # Write n-1 as 2^r * d
    r, d = 0, n - 1
    while d % 2 == 0:
        r += 1
        d //= 2

    # Witness loop
    for _ in range(k):
        a = random.randrange(2, n - 1)
        x = pow(a, d, n)  # a^d % n
        if x == 1 or x == n - 1:
            continue
        for _ in range(r - 1):
            x = pow(x, 2, n)
            if x == n - 1:
                break
        else:
            return False
    return True

Observation: Look at the result that we got when started with basic prompt and refined the result to suit our expectations and when we sent a complex two sentence prompt, the final output is identical (except for function name).

Takeaway

Old belief: You must write perfect prompts, always with relevant context.

New reality: AI is smarter by default. Even basic prompts are solid, and you can add context iteratively to get desired results.

True superpower: Iterating like you would with a human team mate — refine, adjust, collaborate.

So do not stress too much over “prompt hacks.” Start simple, then iterate.

Real AI vs. Automation: What Developers Need to Know

rishisingh007 — Sat, 02 Aug 2025 07:16:00 +0000

As AI continues to dominate headlines, more and more products are marketing themselves as “AI-powered” — from productivity tools to CRMs, chatbots to ecommerce platforms. But when you look under the hood, many of these so-called “AI features” are little more than automated workflows or scheduled scripts.
So it is the time that we ask the hard question: What really counts as AI, and what is just an automation?

The Overuse of “AI”
Today, you’ll find companies claiming to use AI for things like:

Sending a discount offer when sales drop below a threshold.
Triggering an alert when stock is low or creating Purchase order.
Auto-sending a WhatsApp message after a payment fails.

These are useful features, but they are not artificial intelligence. They are plain IF-THEN rules in a computer program, scheduled logic/cron-jobs, or filters that run at set times or conditions.

This is automation. And there’s nothing wrong with it, but we shall not mislabel it as “AI” to ride the hype.

What Is Real AI?

True AI, especially the kind that powers modern breakthroughs, includes technologies like:

LLMs (Large Language Models) : like GPT, Claude, or Mistral
Generative AI : text, image, video, audio generation
Machine Learning: models that learn from data patterns
Computer Vision : models that interpret visual data

What sets real AI apart?

Real AI does at least one of the following:

Understands messy, unstructured input
Learns or adapts from new patterns
Generates context-aware output
Infers new knowledge insights into data based on interrelationship
Perform data summarization

Automation vs AI: A Clear Comparison

Automation is powerful, reliable, fast, and rule-based. But AI is much beyond these rules. For instance, it can reason, infer, summarise and adapt.

Why It Matters

Calling everything AI is more than just misleading as it affects:

User trust: When people expect intelligence and see simple triggers, trust is lost.
Team clarity: Developers, PMs, and stakeholders can misunderstand what’s actually being built.
Funding misuse: Investors may fund startups thinking they’re buying deep tech when it’s just logic wrapped in buzzwords.

If your product uses GPT or other LLMs to summarize contracts, generate content, or answer questions in context — that is real AI. If it just does things on a timer or after a condition is met, that is plain logic. It is useful, but let’s call it what it is.

Where AI Should Be Used

Instead of forcing AI into every feature, let’s focus on where it really adds value:

Document understanding (contracts, policies, research, legal)
Customer service (contextual replies, not just question answer pair responses)
Summarization and analysis of text or numerical data
Personalized content generation
Intelligent search over data

In these areas, AI is not just a marketing layer, it solves the real business problems.

Summary and Final Thoughts

AI is transforming the software landscape, but let us not dilute its meaning. Not every useful feature is intelligent. And not every intelligent type looking tool is using AI, it could just be a plain automation based on plain software logical steps.

We as builders, founders, and developers, we have responsibility to our users and to ourselves, to stay honest and clear.

Let automation do what it is good at, and let AI get it’s due respect where it makes a difference.

Written by : Rishi Pal Singh

Please feel free to share your thoughts on this, as where have you seen “fake AI” being hyped? And where do you think real AI adds the value?

Connect with me on my LinkedIn: www.linkedin.com/in/rishi-singh-b8674751

AI Ticket Summarizer: How to Add AI Features to Your Web or SaaS App (Without Training a Model)

rishisingh007 — Fri, 25 Jul 2025 01:25:23 +0000

Imagine your support dashboard could instantly condense lengthy tickets into sharp summaries — without hiring extra staff or writing any AI models from scratch. Here’s how to make it real.

🔍 The Problem

Customer support teams often deal with long, detailed tickets. Developers and product managers must scan multiple conversations to understand what’s going on. This wastes time and slows decision-making.

What if your system could summarize these tickets using AI, instantly?

✅ The Solution: Plug-in AI Ticket Summarizer API

Using just Node.js + Express, and integrating with OpenAI or OpenRouter, I built a minimalistic, ready-to-use backend that:

🔁 Accepts long ticket text via a POST request

✂️ Returns a short, intelligent summary using gpt-3.5, gpt-4, or mistral

⚙️ Works with either OpenAI or OpenRouter.ai

🔐 API-key protected, portable, and extensible
No frontend? No problem. You can call this API from your own app, chatbot, CRM, or even a spreadsheet tool.

🚀 Quick Demo

Here’s a sample cURL request: (remove back slash and new line while running on command prompt)

curl -X POST http://localhost:3000/summarize \
  -H "Content-Type: application/json" \
  -d '{"ticket": "Hi team, the dashboard takes 30+ seconds to load after the update. Blocking client reporting. Tried cache clearing but didn’t help."}'

And response:

{
"summary": "Dashboard is very slow after update, affecting reporting. Issue confirmed by multiple users."
}

🧩 Tech Stack:

Node.js + Express
OpenAI API or OpenRouter (fallback support)
.env file for secrets
Supports both gpt-3.5-turbo and open-source models like mistral-7b-instruct

🛠 Setup in 2 Minutes

git clone https://github.com/rishisingh007/ai-ticket-summarizer
cd ai-ticket-summarizer
npm install
cp .env.example .env

Then choose either:

Use OpenAI (needs API key)

node index.js

Or use OpenRouter (free options available)

node index-openrouter.js
More setup details in the README: https://github.com/rishisingh007/ai-ticket-summarizer/blob/main/README.md

🧩 How This Works

User submits ticket text
Backend sends it to OpenAI GPT
Prompt asks: “Summarize this ticket briefly”
AI returns a clean summary
Frontend shows the output

🔄 Coming soon:

🌐 React frontend to upload batches of tickets
💾 Export summaries to file or DB
⚠️ Auto fallback if OpenAI quota fails
📬 Slack/email integration for auto-posting

📦 Why This Is Useful

If you’re a developer or team lead:

🎯 Add AI summarization to your support portal or ERP in <10 minutes
🧪 Great for experiments and proof-of-concept demos
🛡️ Fully local, secure, and customizable

👨‍💻 About the Author

I’m Rishi Pal Singh, a Director-level technologist and builder of AI-integrated enterprise tools.

🔗 LinkedIn — https://www.linkedin.com/in/rishi-singh-b8674751

💻 GitHub — rishisingh007