Loading…
Thoughts on data science, ML engineering, and building real projects — written from what I've actually shipped.
The real deployment setup behind this site — Nginx reverse proxy, PM2 process management, Let's Encrypt SSL, and running multiple projects on one EC2 box.
A no-fluff roadmap for learning data science without wasting months on the wrong things first — what matters early, what can wait, and what's mostly noise.
Why I pair a Next.js frontend with a FastAPI backend instead of going all-in on one framework, and how the two talk to each other in a real deployed app.
A practical, no-hype roadmap for building your first real GenAI agent — from a single LLM call to a tool-using agent served over a FastAPI backend.
A clear, example-driven walkthrough of Python's object-oriented programming — classes, objects, inheritance, and polymorphism — for anyone moving past basic scripts.
Node.js, Express, MongoDB, and Cloudinary powering a Learning Management System backend — and why documenting the API with Swagger from day one saved real time.
An API-first backend service for parsing resumes — and why extracting structured fields from a PDF is a genuinely underrated, messy data engineering problem.
A focused Generative AI project built around applying large language model APIs to an actual practical task, rather than just demonstrating that an API call works.
Git tracks your code changes. DVC tracks your data and model changes the same way — here's why that distinction matters once an ML project has more than one contributor or one training run.
A TypeScript + Python data analytics project that takes raw IPL match data and turns it into structured, decision-ready insight — the kind of pipeline real analytics work looks like.
Training a model is the easy part. This project is about the parts around it — preprocessing, evaluation, tracking, and CI/CD — that make a model deployable, not just accurate.
A structured, growing collection of Python data science fundamentals — not a random notes dump, but a curriculum I built for myself and now teach from.
Not every ML project needs to be complex to be worth building well — a FastAPI + scikit-learn + Next.js app that predicts scores from study hours, with a real UI instead of a notebook.
A quiz platform built with Next.js and NestJS where questions can come from a teacher or from an LLM — and both need to fit the same data model.
How retrieval-augmented generation actually works in practice — chunking, embeddings, vector search, and streaming — built as a chatbot that answers questions about my own portfolio.
Employee management, student enrollment, media uploads, and real-time chat — all inside one modular Node.js API. Here's how keeping it modular kept it maintainable.
Building a full-stack HMS with RBAC, AI diagnostics, and a live bed matrix — and why starting from "who uses this" instead of "what does this do" changes the whole architecture.
Chat, voice, and document intelligence in one tutoring platform — the design decisions behind combining three very different input modes into one coherent assistant.
Building a full ANPR pipeline with YOLOv8, EasyOCR, and Gemini — and why the hard part isn't detecting the plate, it's reading it correctly when the lighting is against you.
Why I built a real-time 3D engine for visualizing 110+ data structures and algorithms, and what it took to make sorting, trees, and graphs feel spatial instead of abstract.