Bhavya — Data Scientist

01.

About

I'm a passionate data science undergraduate skilled in Python and AI/LLM development. I thrive on extracting insights from large datasets and enjoy collaborating with teams to deliver impactful, data-driven solutions. My experience in mentoring and technical guidance fosters a collaborative environment that drives success.

02.

Experience

Jun 2026 – Jul 2026

Data Science Intern · Zappify

Developed and deployed comprehensive data science workflows, encompassing data preprocessing, exploratory data analysis (EDA), and visualization utilizing Python, Pandas, and Seaborn.
Implemented and assessed machine learning models using Scikit-learn, applying standard classification metrics such as accuracy, precision, recall, and F1-score.

Nov 2025 – Present

Creative Head · The Empirical Society

Led a team of 10 designers for college technical events, enhancing social media engagement by approximately 35% while coordinating end-to-end design workflows across events, marketing, and editorial teams.

03.

Selected work

R

Resume Screening: ClassifierCheck out

Python · Scikit-learn · TF-IDF · NLTK · GridSearchCV

Constructed an NLP pipeline to classify 962 resumes into 25 categories using text cleaning and TF-IDF with bigrams (20,000 features).
Benchmarked 6 classifiers under a stratified 80/20 split; Linear SVM tuned via 3-fold GridSearchCV (C {0.01–10}) achieved optimal performance — results interpreted with caution given small test set (n=193); exported predictions to test_predictions.csv for auditability.

Python · Scikit-learn · TF-IDF · NLTK · GridSearchCV

GitHub

M

Message Spam FilteringCheck out

Python · Scikit-learn · TF-IDF · Streamlit · Deployed Web App Link

Developed and deployed an NLP classifier on 5,572 SMS messages; transitioned from MultinomialNB to ComplementNB to address 87/13 class imbalance, enhancing spam recall from 72% to 89% at 97.4% test accuracy.
Engineered bigram TF-IDF features (ngram_range=(1,2)) to capture phrase-level spam patterns overlooked by unigrams; applied stratified train/test split with full-data refit before deployment.

Python · Scikit-learn · TF-IDF · Streamlit · Deployed Web App Link

Live Demo GitHub

C

Customer Churn PredictionCheck out

Python · Scikit-learn · StandardScaler · K-Fold CV

Developed a classification pipeline on telecom churn data; benchmarked Logistic Regression, Random Forest, and SVM under K-Fold CV with 80/20 class imbalance managed via class_weight='balanced' on Logistic Regression.
Achieved best model at 81% churn recall compared to 0% for a naive majority-class baseline scoring 83% accuracy — framed recall, not accuracy, as the critical metric for retention decisions.

Python · Scikit-learn · StandardScaler · K-Fold CV

GitHub

A

Airbnb Rental Price PredictionCheck out

Python · Scikit-learn · Pandas · GridSearchCV

Executed an end-to-end regression pipeline on 27,379 NYC listings; compared Linear Regression vs Random Forest (5-fold CV: 44.6% vs 54.5% R²); diagnosed ~100% train / 56.2% test R² gap as a feature ceiling due to missing key variables.
Eliminated price outliers via IQR (25,720 rows), applied LabelEncoder for tree-compatible encoding, and utilized hexbin actual-vs-predicted plots instead of scatter to manage 25K-point overplotting.

Python · Scikit-learn · Pandas · GridSearchCV

GitHub

B

Barbie – Voice-Activated AI AssistantCheck out

Groq LLaMA 3.3 70B · gTTS · NewsAPI · PyWhatKit

Built an end-to-end agentic AI voice assistant with real-time speech-to-text and LLM-powered NLP; integrated autonomous web control (YouTube, live news), achieving sub-3s round-trip interaction latency.
Implemented persona-driven prompt engineering maintaining consistent AI character behaviour across 50+ distinct
conversation scenarios; integrated live NewsAPI and autonomous task execution — directly analogous to AI Copilot and Agentic AI architectures.

Groq LLaMA 3.3 70B · gTTS · NewsAPI · PyWhatKit

GitHub

A

Automated AI ChatbotCheck out

Python · Groq API · LLaMA 3.3 · Prompt Engineering · LLM Integration

Engineered a production-ready conversational AI system using Groq-hosted LLaMA 3.3 70B, implementing structured prompt engineering pipelines with multi-turn dialogue management across 10+ exchange threads without memory degradation.
Designed async LLM inference architecture achieving sub-2s average response latency; built modular pipeline supporting plug-and-play LLM backend swapping — demonstrating production-grade system design thinking.
Implemented automated fallback and error-recovery logic; reduced hallucination rate ~40% through iterative output validation and structured response formatting.

Python · Groq API · LLaMA 3.3 · Prompt Engineering · LLM Integration

GitHub

04.

Skills & tools

LanguagesAI/MLFrameworks/LibrariesDatabasesTools & PlatformsOther Technical Skills

05.

Education

Guru Gobind Singh Indraprastha University – GTB4CEC

B.Tech – Computer Science & Engineering (Data Science)

2024 – 2028 · CGPA: 8.90 / 10

06.

Leadership & activities

Nov 2025 – Present

Creative Head · The Empirical Society

07.

Certifications

GenAI-Powered Data Analytics

Tata iQ · 2026

Data Visualization

TCS · 2026

08.

Get in touch

Have a role or a project in mind? I'd love to hear from you.

bhavyakumar.2511@gmail.com +91 9773559908