Pigeonic (2026)

14/01/2026

🎯 Course Completed with Success!

I’m excited to share that I’ve successfully completed the AI Engineering Bootcamp for Programmers with an overall score of 94.3%. 🚀

This journey strengthened my understanding of AI engineering concepts, practical problem-solving, and real-world implementation through live classes, assignments, quizzes, and assessments.

Grateful for the learning experience and looking forward to applying these skills in real projects and future challenges. 💡🤖

24/12/2025

I’m pleased to share that I have successfully completed the AI Engineering Bootcamp for Programmers from Ostad.

This structured and hands-on program strengthened my practical understanding of Machine Learning, Deep Learning, NLP, and AI Engineering, with a strong focus on real-world applications and deployment.

Special thanks to my instructor, Tahmid Rahman, for his clear guidance, consistent support, and valuable mentorship throughout the course.

Key learning areas included:
Machine Learning pipelines, regression and classification models, unsupervised learning, deep learning, computer vision, transformers, NLP, prompt engineering, RAG systems, AI agents, and model deployment using FastAPI, Docker, and MLflow.

I also worked on multiple applied projects, including a Bangla RAG system and a customer support chatbot with Redis-based memory, focusing on scalable and production-ready AI solutions.

Looking forward to applying these skills to real-world problems and continuing my learning journey.

26/11/2025

Python, React JS, Loveable, Vercel, GitHub, CI/CD, Docker, Render, FastAPI, n8n, Webhook, Webscrapping, Using LLMs, Google Sheet Save, Sending Mail, HTTP Request — all working together in a single full-stack automated workflow.

The posted video is showcasing how I built an AI-powered Article Analyzer App using a fully automated pipeline where both the React JS frontend and the Python FastAPI backend were generated with single-prompt AI scaffolding and deployed through GitHub integrations.

Here is the high-level workflow:

- React JS frontend generated in Loveable with one prompt, then slightly modified and deployed on Vercel via GitHub CI/CD.

- Python FastAPI backend generated with a single LLM prompt, lightly refined, Docker-ready, and deployed on Render connected to GitHub.

- n8n workflow receives a Webhook request containing email + URL from the frontend.

- n8n uses Firecrawl Webscrapping to fetch page content.

- Two Gemini LLM models running one after one: one summarizing the article, another extracting 2-3 insights.

- n8n saves all processed data into a 2-factor authenticated Google Sheet.

- The same data is emailed to the user based on the original webhook input.

- n8n sends the final result back to FastAPI via HTTP Request, which updates the status.

- Frontend keeps polling FastAPI until success status arrives, showing a processing loader meanwhile.

- Once success is confirmed, the final summarized results and insights appear instantly on the frontend UI.

I will share the full code, prompts, architecture, and GitHub repos in a detailed future post. App Link in comment. Try out!!

19/11/2025

Data may look chaotic, but hidden inside it are groups, patterns, behaviors, and insights waiting to be discovered.

Unsupervised Machine Learning, K-Means Clustering, the Elbow Method, and Principal Component Analysis help us understand unknown data without labels. These tools reveal hidden patterns, reduce complexity, and convert confusion into clarity.

Imagine standing in a room full of strangers. Without asking anyone, you observe who talks to whom, who behaves similarly, and who shares common interests. Eventually, you can identify groups. That is exactly how Unsupervised Learning works.

K-Means takes this further by dividing data into K groups based on similarity. It starts with random centers, groups data, updates the centers, and repeats until stable patterns emerge. The Elbow Method helps decide how many groups are meaningful by identifying the point where adding new clusters no longer reduces error significantly.

Principal Component Analysis solves a different problem. When data has too many features, PCA acts like a filter that keeps only the strongest patterns. It reduces dimensions while keeping most information intact. This helps in visualization, compression, and improving model performance.

These methods solve real problems such as customer segmentation, fraud pattern detection, document grouping, market targeting, image organization, and noise reduction in high-dimensional datasets.

Their strengths lie in simplicity, speed, pattern discovery, and preprocessing capability. Weaknesses include sensitivity to noise, difficulty interpreting components, and challenges evaluating performance without labeled data.

In my Medium article, I explained all topics with step-by-step analogies, math formulas, real-life examples, strengths, weaknesses, and evaluation methods.

I have written a detailed article in Medium with full explanations, real examples, and required codes. Check link in first comment.

30/10/2025

Building AI APIs isn’t just about predictions, it’s about taking a model from your notebook to the real world.

Recently, I developed and deployed a Heart Disease Prediction API using Python, FastAPI, and Docker, hosted for free on Render.

I trained a RandomForest model using the Heart Disease dataset from Kaggle, saved it with joblib, created an end-to-end FastAPI backend, validated requests using Pydantic, containerized it, tested locally with Docker Compose, and finally deployed it live.

This project demonstrates the full journey, data preprocessing, model serving, containerization, and cloud deployment.

I also explored caching techniques using Redis (commented in code for demo), showing how we could reduce server load for repeated queries.
Everything is open-sourced and well-documented, including full scripts and explanations.

I’ve written a detailed article on Medium with all steps, commands, and explanations.

Check the link in the first comment for the full write-up and GitHub repo.

24/10/2025

Frontend Animation — Framer Motion vs GSAP: when to use each and 20 real examples.

I recently audited several product UIs and found the same problem: animations were either missing or inconsistent — harming clarity more than helping it. As a frontend developer, I use Framer Motion for declarative UI transitions and GSAP for cinematic scenes. This combination lets teams ship polished micro-interactions quickly while ramping up creativity when required.

In practice, I split responsibilities: use Framer Motion for buttons, modals, list reorders, layout shifts, and small scroll reveals — because it integrates directly with React, supports layout animation, and is easy to maintain. Use GSAP for advanced timelines, complex SVG morphs, motion paths, and scroll-triggered pinning when you need precise control over sequence, easing, and timing.

Here’s a list of the 20 animations I implemented to demonstrate these techniques:

Framer Motion Examples (10)
1. Fade In
2. Hover
3. Stagger List
4. Layout Shift
5. Modal Backdrop
6. Page Transition
7. Rotate
8. Scale Up
9. Scroll Parallax
10. Shape Morphing / Slide Up / Text Typing

GSAP Examples (10)
1. Fade Slide
2. ScrollTrigger Reveal
3. Timeline
4. Motion Path
5. Shape Morphing
6. Cursor Follow
7. Pinning
8. ClipPath
9. SVG Stroke
10. Complex Sequence

View all animation examples with codes at https://animate.pigeonic.com

Each animation was built with React + TypeScript + Tailwind CSS, and I included line-by-line code explanations. Framer Motion covers UI-friendly transitions like hover effects, modals, and list animations. GSAP handles creative, cinematic, and scroll-driven effects such as morphing SVGs, timeline sequencing, or magnetic cursor interactions.

If you’re hiring for front-end, animation experience matters — it separates “works” from “delightful.” Product managers and recruiters should ask candidates how they balance accessibility, performance, and animation complexity when designing interfaces.

I have published a detailed article on Medium covering all 20 examples with production-ready snippets for hero transitions, parallax storytelling, and interactive elements. Check the link in the first comment.

21/10/2025

Can machine learning really help predict diabetes before it happens?

I built a complete end-to-end Supervised Machine Learning Classification Project that predicts diabetes outcomes using the Pima Indians Diabetes Dataset.

From data loading, EDA, scaling, model training, and evaluation, I tested four algorithms — Logistic Regression, KNN, Decision Tree, and Random Forest.

I’m happy to share this End-to-End Diabetes Prediction with Machine Learning— Logistic Regression, KNN, Decision Tree, and Random Forest Explained Step-by-Step project I worked on. Check it out here: https://lnkd.in/g_unx_zj

Each model was trained on 80% of the dataset and tested on 20%. After feature scaling, models were evaluated with accuracy score, confusion matrix, and ROC-AUC.

Key Highlights:
- EDA: Checked nulls, distribution, outcome balance, and visualized histograms.
- Scaling: Standardized features for fair model comparison.
- Models: Compared accuracy and ROC curves — Random Forest performed best.
- Visualization: Plotted Decision Tree, KNN error rate, and accuracy bar charts.

Real-life applications include:
- Predicting health risks early
- Reducing diagnosis time
- Empowering doctors with data-backed insights

This project shows how machine learning in healthcare can make a difference in people’s lives.

I’ve written a detailed Medium article with step-by-step explanations, code snippets, and real use cases.

Check the link in the first comment.

16/10/2025

Artificial Intelligence (AI), Deep Learning, Computer Vision, Super Vision, Object Tracking.

Just completed a People Counting & Heatmap project! Built a Python system using YOLOv8 & ByteTrack to detect and track people in video, count IN/OUT movements across lines, and generate cumulative heatmaps. Input/output videos ready.

Detailed article with full script coming soon.

16/10/2025

React to Next JS, Vue to Nuxt JS – When and Why to Move

The modern web moves fast. Developers often start with React or Vue, and then, somewhere between growing SEO needs and performance bottlenecks, realize it’s time to shift gears. That’s where Next.js and Nuxt.js come in.

I’ve seen this transition in real projects: a small startup building a single-page React app later transforms it into a full-scale, SEO-optimized, server-rendered platform with Next.js. The same applies to Vue to Nuxt migrations. The shift isn’t just technical, it’s strategic.

React and Vue are powerful front-end libraries, flexible, fast, and perfect for SPAs or dashboards. But when your app needs SEO, server-side rendering (SSR), static generation (SSG), image optimization, or faster first loads, that’s the moment to move to Next.js or Nuxt.js.

Both frameworks take your app from “good” to “production-grade.”

Next.js builds on React with SSR, routing, API endpoints, image optimization, and incremental static regeneration (ISR).

Nuxt.js enhances Vue with hybrid rendering, automatic routing, and better SEO, all with elegant simplicity.

In short:
Use React/Vue for dynamic dashboards, SPAs, and internal tools.
Migrate to Next.js or Nuxt for SEO-focused sites, e-commerce, blogs, or marketing platforms.
Both ecosystems now support PWAs, caching, TypeScript, and edge rendering. Hosting on Vercel, Netlify, or Cloudflare makes deployment effortless.

If you want to future-proof your stack, start small, scale smart, and evolve strategically.

I’ve written a detailed Medium article with SSR/CSR/SSG explanations, caching, performance tables, tool stacks, and 10+ real-life use cases for each tech.

Check the link in the first comment.

09/10/2025

A Remarkable Milestone in My Journey with Bangladesh Police Budget & Finance Software

Today, I attended a training session at Bangladesh Police Headquarters for a project that has been a major part of my seven-year-old career journey over the last four years, the Budget & Finance Management Software for Bangladesh Police.

It has been a challenging and rewarding experience to see this system grow from concept to reality. The software is a fully customized financial management platform capable of handling large-scale, complex financial data across all police units in Bangladesh.

I’m grateful to serve as a Senior Software Engineer at R3Soft Limited, taking full responsibility for system planning, architecture, database design, core development, deployment, and server management.

Special thanks to Addl DIG Atiqur Rahman, BPM (Finance-1 Section), sir and Addl SP Iqbal Hossain (Finance-1 Section), sir Bangladesh Police Headquarters, for their exceptional leadership and support throughout the project.

Thanks to Rony Mondal, CEO of R3Soft, sir for his trust and guidance.

The system manages:
- All types of budget and distribution records nationwide
- Automatic government letter generation with multiple formats
- Large financial reports (100,000+ rows at once)
- Dynamic entry modules with hundreds of filters
- Over 30 connected modules for all financial operations

The biggest challenges were automating complex government financial rules, processing massive data efficiently, and ensuring system reliability.

Developed using Laravel, jQuery, Bootstrap, and MySQL, this project shows how technology can modernize government finance management.

The journey continues, we’re now scaling it with more users, features, and higher performance to make it a central financial hub for the Bangladesh Police.

Special thanks to teammates Ashik and Uttam for their valuable contributions to the full stack.

08/10/2025

Every AI model tells a story. But the real story begins when we start evaluating it.

Accuracy, confusion matrix, precision, recall, F1, ROC-AUC — these are not just numbers. They’re the DNA of intelligent decisions in AI, ML, and deep learning.

In real-world projects, accuracy alone can lie. A fraud detection model with 98% accuracy might still miss the real frauds — because accuracy ignores imbalance. That’s where precision, recall, and F1-score become our truth indicators.

A confusion matrix shows where predictions go wrong, a ROC curve shows how confident the model is at different thresholds, and AUC summarizes the model’s overall decision power.

Each metric fits a specific model:
- Logistic Regression loves ROC-AUC
- Random Forests thrive on F1
- Neural Networks benefit from Log Loss and Recall

Evaluation is not just about math — it’s about trust.
The moment we skip evaluation, we risk deploying bias, overfitting, and wrong predictions into the world.

And remember — evaluation isn’t only an AI principle. In software engineering, every commit, every test, every release is a form of evaluation.

Because quality without measurement is just assumption.

I’ve written a detailed article in Medium with full explanations, mathematical logics, real-world examples, and Python codes.

07/10/2025

Every JavaScript developer has faced it, a tiny undefined variable breaking a whole feature before launch. I lived through that chaos once in a massive React project. Debugging felt endless, and every fix risked another bug.

That’s when I discovered TypeScript, not a replacement for JavaScript, but an evolution. A superset that adds type safety, interfaces, and compile-time checking to the language we already love.

Imagine driving at night: JavaScript is the open road with no lights. TypeScript turns on the headlights. You see what’s ahead before you crash.

TypeScript helps developers:
- Catch errors before runtime.
- Define clear data contracts with interfaces.
- Collaborate confidently across large teams.
- Refactor safely across hundreds of files.

It’s now used in every serious large-scale app — from fintech dashboards to enterprise portals — because predictability equals productivity.

In my projects, switching to TypeScript reduced bugs by over 60% and made onboarding smoother. New developers could understand structures instantly thanks to interfaces.

TypeScript doesn’t slow you down. It lets you move faster with confidence.

If JavaScript was the wild west, TypeScript is the modern city, structured, safe, and scalable.

I have written a detailed article on this topic with TypeScript code examples.
👉The link is in the first comment.

Pigeonic

14/01/2026

24/12/2025

26/11/2025

19/11/2025

30/10/2025

24/10/2025

21/10/2025

16/10/2025

16/10/2025

09/10/2025

08/10/2025

07/10/2025

Address

Telephone

Website

Alerts

Shortcuts

Share