SmitKothari

Computer Science & Engineering Student | VIT Vellore

Scroll to explore

ABOUT ME

Hi! I’m Smit Kothari, an aspiring Data Analytics and Machine Learning enthusiast who genuinely enjoys making sense of complex problems.

I love building and refining solutions, whether it’s new ML models, digging into data, or just figuring out how things work.
I stay curious, committed to learning, and driven by meaningful challenges.

Experience That Shaped Me

Machine Learning Intern

June-July 2025

Polestar Analytics

Designed and deployed an end-to-end sales volume prediction pipeline using a config-driven architecture, automating 70+ feature combinations and reducing manual forecasting effort by 80%.

Built and optimized multiple forecasting models (ARIMA, Prophet, Random Forest, XGBoost, LightGBM) with advanced feature engineering and Optuna-based hyperparameter tuning, achieving 93.9% accuracy and WMAPE as low as 5.4% across 100+ brand–region–SKU combinations.

PythonMachine LearningData AnalyticsExcel

App Tester & Youth Intern

2019 – 2021

AHC

Collaborated with a US-based NPO focused on adolescent health advocacy by contributing to app functionality and ensuring 100% curriculum alignment with health standards

Optimized cross-platform app performance by 22% through 30+ functional tests while simultaneously revising health education curricula to enhance content precision

Mobile TestingUX DesignQuality Assurance

Skills & Expertise

Programming Languages

Python

Java

C++

HTML/CSS

JavaScript

SQL (MySQL, Oracle)

Frameworks

Flask

MongoDB

Seaborn

Git

MATLAB

Data Science & ML

TensorFlow

Scikit-learn

NLP(Bert models)

PyTorch

Arima

Built With Passion

Trust-Based Serverless Function Selection System

Patent Pending

April 2026

JavaCloudSim PlusDistributed SystemsServerlessNetBeans

A reliability-aware serverless scheduling framework simulated using CloudSim Plus that dynamically routes cloud function invocations using adaptive trust scoring instead of traditional FCFS/load-based allocation.

Designed a reliability-aware serverless scheduling framework that dynamically routes cloud function invocations using adaptive trust scoring instead of traditional FCFS/load-based allocation.

Developed a mathematically grounded trust engine combining Bayesian reliability estimation, EWMA recency scoring, latency normalization, cold-start penalization, and failure streak analysis for self-learning invocation selection.

Validated the system through CloudSim Plus simulations across workloads up to 3,000 invocations, achieving up to 12.99% higher success rates and over 50% reduction in retries with respect to FCFS baselines.

Stickles

WebsiteJanuary 2026

ReactNext.JSSupabase (PostgreSQL)Tailwind CSS

A full-stack e-commerce platform built with Next.js and Supabase featuring real-time product discovery, cart & order workflows, and optimized media delivery.

Architected and deployed a full-stack e-commerce platform with real-time product discovery, cart & order workflows, sustaining sub-300ms API response times under concurrent load.

Built scalable backend services using Supabase (PostgreSQL) with row-level security and JWT authentication, optimizing queries and integrating Cloudinary for faster media upload and delivery, resulting in ~35-40% lower latency and reliable cloud-native deployment.

GitHub

Bradley-Terry Argument Ranking Framework

March 2026

PythonPyTorchNLPDeBERTa-v3

A topic-aware NLP argument ranking framework using DeBERTa-v3-small and Bradley-Terry probabilistic modelling, achieving 92.91% classification accuracy and 0.9291 weighted F1-score on the IBM ArgQ-9.1kPairs benchmark.

Developed a topic-aware NLP argument ranking framework using DeBERTa-v3-small and Bradley-Terry probabilistic modelling, achieving 92.91% classification accuracy and 0.9291 weighted F1-score on the IBM ArgQ-9.1kPairs benchmark.

Engineered a dual forward-pass architecture with order-flip augmentation, asymmetry regularization (λ=0.3), and Monte Carlo Dropout inference, improving positional robustness and enforcing tournament consistency across 8,482 augmented training samples.

Outperformed prior DeBERTa-based baselines by +11.74 percentage points, achieving up to 95.83% per-topic accuracy while providing calibrated uncertainty estimation through 10 stochastic inference passes for real-time debate analytics and explainable AI evaluation.

GitHub Research Paper

EventEase

Website

FlaskMongoDBJavaScriptHTML/CSSPython

Flask + MongoDB based Event Management Website for university students and clubs and chapters. Boosted event discovery by 5x with dynamic filters and real-time updates.

Dedicated accounts for university clubs/chapters to post and manage events.

Categorization of events (Educational, Cultural, Sports) for enhanced discoverability.

Students can browse and RSVP for events; provides clubs with real-time attendee details list

GitHub

My Academic Journey

Vellore Institute of Technology

B.Tech in Computer Science & Engineering (sp. in Business Systems)

Aug 2023 – July 2027

CGPA: 9.15

Udayachal High School

Secondary School Certificate (SSC)

June 2015 – Apr 2021

Percentage: 95.2%

Let's Create Something Amazing Together!

Ready to collaborate on exciting projects or discuss innovative ideas? I'm always open to new opportunities and connections.

Phone

+91 98190 48380

Personal Email

smit.kothari0305@gmail.com

University Email

smit.kothari2023@vitstudent.ac.in

GitHub

github.com/smk355

linkedin.com/in/smit-kothari-2977651b7

LeetCode

leetcode.com/u/eBiFFrR0Q8