PinkeshOS v2.0

about.me — Pinkesh Darji

Pinkesh Darji

Builder · Analyst · AI Enthusiast

Full Stack Dev Data Analyst Open to Work Vibe Coder

Bio

Education

Interests

Who Final year MCA student at CHARUSAT University who builds software people actually use. I enjoy working at the intersection of engineering, data, and AI.

Currently Interning at Vezapp Solutions LLP — building data pipelines, SQL architectures, and production dashboards for digitalized foundry logbooks.

Interests Software Engineering · Data Science · Artificial Intelligence · Vibe Coding with music on 🎵

Looking for Full-time roles in Data Science & Software Engineering

Location Borsad, Gujarat, India 🇮🇳

Master of Computer Applications (MCA)

CHARUSAT University

2024 – 2026 · Pursuing

BSc in Computer Applications & IT

Sardar Patel University (SP University)

2021 – 2024 · Completed

Cert Google Data Analytics Professional Certificate — Data Cleaning, Visualization, Analysis

Cricket

Music

Tech YouTube

AI / ML Research

Vibe Coding

Data Stories

Vibe I build best with lo-fi beats or a cricket match in the background. The kind of dev who goes down rabbit holes at 2am because something was interesting, not just required.

bash — pinkesh@portfolio ~ zsh 5.9

pinkesh@portfolio:~$

Projects — 5 apps shipped

🚀

Elevo

AI Placement Prep Platform · Full-Stack Web App

Live Demo GitHub

Overview

Elevo is an AI-powered placement preparation platform built to bridge the gap between academic learning and industry readiness. It combines coding practice, aptitude prep, resume-based mock interviews, tutor-led learning, analytics, and performance tracking into one structured system for both individual learners and organizations such as colleges or training institutes.

Core Features

Unified preparation workflow — coding practice, aptitude tests, resume analysis, AI mock interviews, and readiness tracking in one platform instead of multiple disconnected tools.
Role-based access control — supports students, tutors, admins, organization members, and institutional use cases with separate permissions and dashboards.
Resume-based interview engine — analyzes uploaded resumes and generates tailored technical or HR interviews with AI feedback.
Social learning and tutor content — includes a learning feed, tutor-managed material, and guided preparation resources.
Analytics + subscription features — tracks performance trends, learner progress, and gated premium access for scalable rollout.
Live coding practice — supports remote code execution and test-case validation for programming rounds.

Hardest Challenge

🔥 Platform orchestration — The main challenge was designing one cohesive system where AI interviews, resume parsing, role-based flows, coding execution, analytics, and subscription logic all work together cleanly inside the same product without feeling fragmented.

Tech Stack

PythonDjango 5PostgreSQLDjango REST FrameworkGemini APIOpenAI APIJDoodle APIDjango-Q2Tailwind CSSBootstrap FormsWhiteNoiseGunicornRenderpdfplumberdocx2txt

IPL Win Probability Predictor

Sports Analytics · Machine Learning Product

Live Demo GitHub

Tagline

Because cricket matches are won in the data, before they're won on the field.

Overview

What if you could quantify pressure instead of just feel it? This project is a live ML-powered web app that reads the current state of a T20 match — score, overs, wickets, target, and chase tempo — and predicts in real time who is winning and by how much. Not a vibe. A number. With a reason behind it.

The Problem

Cricket analysts and fans instinctively know when a team is under pressure, but gut feel does not scale. The goal was to turn match intuition into a reproducible, explainable machine learning system, then wrap it in an interface clean enough to belong on a broadcast screen.

Feature Engineering

Pressure Index (RRR - CRR) — one number that captures how hard the chase really is. Positive means the batting side is behind the curve. Negative means cruising.
Required Run Rate vs Current Run Rate — models the two speeds of the innings and the gap between them.
Innings Progress (0 to 1) — gives the model match phase awareness, because death overs change everything.
Wickets Left — framed as remaining resources, not just collapse history.

Insight — once these features were engineered, the problem became close to linearly separable. Better features beat fancier models here.

Model Building

Model	Accuracy	ROC-AUC
Logistic Regression	97.6%	0.9888
Random Forest	97.5%	0.9888
Gradient Boosting	97.5%	0.9884

Pipeline — trained on 44,000+ synthetic IPL-style match snapshots, validated with 5-fold StratifiedKFold to handle class imbalance, then packaged as ColumnTransformer → StandardScaler + OneHotEncoder → Model. Serialized with Joblib.

UI & Product Design

Dark sports-analytics UI — minimal layout with all signal and no clutter.
Live win % display — giant probability readout with color-coded confidence.
Split probability bar — both teams shown side by side, not buried in a single stat.
Contextual insight text — translates match state into plain English.
What-If simulator — lets users test scenarios like a 15-run over and see instant updates.
Probability curve + momentum badges — turns changing score states into readable product signals.

Deployment

Deployed on Streamlit Cloud with zero infrastructure overhead. One URL. Any device. Immediate access.

Tech Stack

PythonScikit-learnStreamlitPlotlyPandasNumPyJoblibCustom CSS

Takeaway

End-to-end thinking — This project is not just about cricket. It shows the ability to take a messy real-world problem, engineer the right features, choose the right model instead of the fanciest one, and ship the result as a polished usable product.

HouseVue

ML-Powered House Price Prediction App

Live Demo GitHub

Overview

HouseVue is an end-to-end machine learning web application that predicts residential property prices based on key house features such as area, bedrooms, bathrooms, floors, condition, year built, and location. The project combines data preprocessing, model training, hyperparameter tuning, and an interactive Streamlit interface to deliver real-time price estimates with performance insights.

Detailed Description

This project was built to demonstrate the complete workflow of a machine learning application, from data preparation to deployment. I developed a regression-based prediction system using Python and Scikit-Learn, where multiple models were benchmarked to identify the best-performing approach. The final solution was integrated into a Streamlit web app that allows users to enter property details and instantly receive predicted house prices.

Product depth — I also added model performance metrics, feature importance analysis, sensitivity insights, and market overview sections. The project was prepared for deployment with a production-ready structure, dependency management, and GitHub integration.

Core Features

Predicts house prices from user-input property details.
Benchmarks multiple regression models to select the best performer.
Uses preprocessing pipelines for cleaner and more reliable predictions.
Applies hyperparameter tuning to improve model accuracy.
Shows feature importance and model insights.
Provides an interactive and deployment-ready Streamlit UI.

What I Learned

How to build and deploy an end-to-end machine learning project.
How to preprocess structured data using Scikit-Learn pipelines.
How to compare and tune regression models for better performance.
How to turn an ML model into an interactive web application using Streamlit.
How to structure a project for GitHub and deployment platforms.

Tech Stack

PythonScikit-LearnStreamlitPandasNumPyMatplotlibJoblibGitGitHub

skills — tech stack

Frontend

React

Angular

JavaScript

Tailwind

HTML/CSS

Backend

Python

Django

Django REST

Node.js

REST APIs

MongoDB

PostgreSQL

Data & Analytics

Power BI

Looker

SQL

Streamlit

Scikit-Learn

Pandas

ML / AI

LLM APIs

Git/GitHub

work history

Currently Active

Intern — Data & Engineering

Vezapp Solutions LLP

Dec 2025 – Present · Surat, Gujarat

Designing and maintaining SQL table schemas for digitalized foundry logbook applications — replacing manual paper-based data entry with structured relational databases.
Syncing SQL tables with client-facing applications to ensure real-time data accuracy across the pipeline.
Building production-grade interactive dashboards in Looker Studio and Power BI for operational analytics and reporting.
Contributing to full-stack feature development in the core application platform.

SQLLooker StudioPower BIData PipelinesFull Stack

Self-Directed Engineering

Academic & Personal Projects

2021 – Present · BSc → MCA

Built 8+ full-stack applications across BSc and MCA — from wellness platforms to AI career prep tools. Each project chosen to solve a real problem, not just complete an assignment.

MEAN StackDjangoAI/MLData Analytics

stats.dashboard — Pinkesh Darji

Projects

5+

shipped apps

Coding

4+

years active

Stack

15+

technologies

Status