Vaishnavi's Portfolio

Master’s in Data Science ,

University of Arizona (May 2026)

2+ years building scalable ETL pipelines and optimizing SQL workloads.

I turn raw data into decisions and language into intelligence — building ML systems, RAG pipelines, and dashboards that don't just report, they reveal.

Master’s in Data Science ,

University of Arizona (May 2026)

2+ years building scalable ETL pipelines and optimizing SQL workloads.

Open to Work

Contact me

Vaishnavi Kulkarni

Open to Work

Vaishnavi Kulkarni

Data / AI-ML

Contact me

@ SELECTED WORK

YouTube Multilingual QA System

LangChain

Llsms-3.1

FAISS

BM25

Groq

Built a hybrid RAG system that makes any YouTube video instantly queryable , in any language — using semantic + keyword search with sub-10-second inference.

35 % Retrieval accuracy gain

YouTube Multilingual QA System

LangChain

Llsms-3.1

FAISS

BM25

Groq

Built a hybrid RAG system that makes any YouTube video instantly queryable , in any language — using semantic + keyword search with sub-10-second inference.

35 % Retrieval accuracy gain

Enterprise Document Intelligence Assistant

LangChain

RAG

LSTM

Streamlit

Deployed a private, offline document Q&A system with cited answers , built for enterprises where data never leaves the building.

35 % Retrieval accuracy gain

Enterprise Document Intelligence Assistant

LangChain

RAG

LSTM

Streamlit

Deployed a private, offline document Q&A system with cited answers , built for enterprises where data never leaves the building.

35 % Retrieval accuracy gain

Telco Customer Churn Prediction

XGBoost

MLflow

FastAPI

Docker

Github Action

End-to-end MLOps pipeline that flags at-risk customers before they churn shipped as a production REST API with automated CI/CD.

35 % Retrieval accuracy gain

Telco Customer Churn Prediction

XGBoost

MLflow

FastAPI

Docker

Github Action

End-to-end MLOps pipeline that flags at-risk customers before they churn shipped as a production REST API with automated CI/CD.

35 % Retrieval accuracy gain

Opioid Overdose Mortality Prediction

EDA

Time Serie Analysis

Python

PowerBI

0.85. Modeled 20 years of federal mortality data to predict overdose hotspots by state and demographic turning CDC records into an actionable public health targeting tool.

35 % Retrieval accuracy gain

Opioid Overdose Mortality Prediction

EDA

Time Serie Analysis

Python

PowerBI

0.85. Modeled 20 years of federal mortality data to predict overdose hotspots by state and demographic turning CDC records into an actionable public health targeting tool.

35 % Retrieval accuracy gain

Emotion Classification

LSTM

NLP

BERT

Built a deep NLP system that detects co-occurring emotions in text because real human language is never just positive or negative.

35 % Retrieval accuracy gain

Emotion Classification

LSTM

NLP

BERT

Built a deep NLP system that detects co-occurring emotions in text because real human language is never just positive or negative.

35 % Retrieval accuracy gain

Customer Retention & Sales Analytics Dashboard

PowerBI

PosterSQL

Python

RFM

Cohort Analysis

End-to-end Power BI dashboard over PostgreSQL and Python — cohort analysis, RFM segmentation, and revenue trends surfacing high-value segments

35 % Retrieval accuracy gain

Customer Retention & Sales Analytics Dashboard

PowerBI

PosterSQL

Python

RFM

Cohort Analysis

End-to-end Power BI dashboard over PostgreSQL and Python — cohort analysis, RFM segmentation, and revenue trends surfacing high-value segments

35 % Retrieval accuracy gain

Agentic Energy Anomaly Detection System

LangChain

LSTM

RAG

Streamlit

AI Agents

4-agent LLM orchestration system with hybrid RAG pipeline and LSTM autoencoder trained on NASA battery cycling data

35 % Retrieval accuracy gain

Agentic Energy Anomaly Detection System

LangChain

LSTM

RAG

Streamlit

AI Agents

4-agent LLM orchestration system with hybrid RAG pipeline and LSTM autoencoder trained on NASA battery cycling data

35 % Retrieval accuracy gain

Housing Affordability

Snowflake

dbt

Power BI

SQL

ETL Pipeline

End-to-end ELT pipeline ingesting Zillow, FRED & BLS datasets into Snowflake across 200+ US metros and 10 years of records

35 % Retrieval accuracy gain

Housing Affordability

Snowflake

dbt

Power BI

SQL

ETL Pipeline

End-to-end ELT pipeline ingesting Zillow, FRED & BLS datasets into Snowflake across 200+ US metros and 10 years of records

35 % Retrieval accuracy gain

Housing Affordability

Snowflake

dbt

Power BI

SQL

ETL Pipeline

End-to-end ELT pipeline ingesting Zillow, FRED & BLS datasets into Snowflake across 200+ US metros and 10 years of records

35 % Retrieval accuracy gain

Suspicious URL Predictor

FastText

APIs

Streamlit

Docker

NLP

Hybrid URL threat detection pipeline combining FastText NLP and heuristic scoring with graph-based campaign clustering — live scored via a containerized Streamlit dashboard.

35 % Retrieval accuracy gain

Suspicious URL Predictor

FastText

APIs

Streamlit

Docker

NLP

Hybrid URL threat detection pipeline combining FastText NLP and heuristic scoring with graph-based campaign clustering — live scored via a containerized Streamlit dashboard.

35 % Retrieval accuracy gain

Automotive Auction Analytics

SQL

dbt

Power Bi

Python

ELT Pipeline

Analyzed 50K wholesale auction transactions to surface structural buying opportunities and regional arbitrage across $322M in Ford F-150 auction data.

35 % Retrieval accuracy gain

Automotive Auction Analytics

SQL

dbt

Power Bi

Python

ELT Pipeline

Analyzed 50K wholesale auction transactions to surface structural buying opportunities and regional arbitrage across $322M in Ford F-150 auction data.

35 % Retrieval accuracy gain

@ SELECTED WORK

YouTube Multilingual QA System

LangChain

Llsms-3.1

FAISS

BM25

Groq

Built a hybrid RAG system that makes any YouTube video instantly queryable , in any language — using semantic + keyword search with sub-10-second inference.

35 % Retrieval accuracy gain

Enterprise Document Intelligence Assistant

LangChain

RAG

LSTM

Streamlit

Deployed a private, offline document Q&A system with cited answers , built for enterprises where data never leaves the building.

35 % Retrieval accuracy gain

Telco Customer Churn Prediction

XGBoost

MLflow

FastAPI

Docker

Github Action

End-to-end MLOps pipeline that flags at-risk customers before they churn shipped as a production REST API with automated CI/CD.

35 % Retrieval accuracy gain

Opioid Overdose Mortality Prediction

EDA

Time Serie Analysis

Python

PowerBI

0.85. Modeled 20 years of federal mortality data to predict overdose hotspots by state and demographic turning CDC records into an actionable public health targeting tool.

35 % Retrieval accuracy gain

Emotion Classification

LSTM

NLP

BERT

Built a deep NLP system that detects co-occurring emotions in text because real human language is never just positive or negative.

35 % Retrieval accuracy gain

Customer Retention & Sales Analytics Dashboard

PowerBI

PosterSQL

Python

RFM

Cohort Analysis

End-to-end Power BI dashboard over PostgreSQL and Python — cohort analysis, RFM segmentation, and revenue trends surfacing high-value segments

35 % Retrieval accuracy gain

Agentic Energy Anomaly Detection System

LangChain

LSTM

RAG

Streamlit

AI Agents

4-agent LLM orchestration system with hybrid RAG pipeline and LSTM autoencoder trained on NASA battery cycling data

35 % Retrieval accuracy gain

Housing Affordability

Snowflake

dbt

Power BI

SQL

ETL Pipeline

End-to-end ELT pipeline ingesting Zillow, FRED & BLS datasets into Snowflake across 200+ US metros and 10 years of records

35 % Retrieval accuracy gain

Suspicious URL Predictor

FastText

APIs

Streamlit

Docker

NLP

Hybrid URL threat detection pipeline combining FastText NLP and heuristic scoring with graph-based campaign clustering — live scored via a containerized Streamlit dashboard.

35 % Retrieval accuracy gain

Automotive Auction Analytics

SQL

dbt

Power Bi

Python

ELT Pipeline

Analyzed 50K wholesale auction transactions to surface structural buying opportunities and regional arbitrage across $322M in Ford F-150 auction data.

35 % Retrieval accuracy gain

Made from the scratch

@ ABOUT

I have been working at the intersection of data, business, and product, turning messy information into clear insights that help teams make better decisions.

Resume

That’s honestly where I do my best work. With 2+ Years of experience in data analytics, problem-solving, and cross-functional collaboration, I’ve learned to move from raw data to meaningful stories without losing sight of the why.

Resume

@ SKILLS & TOOLS

Languages & Tools

Python

SQl

Tableau

Excel

Git

Jupyter

Google Colab

NoSQL

Power BI

PySpark

Analytics & Statistics

EDA

Time-Series

ARIMA

A/B Testing

Regression

Statistical Modelling

Hypothesis Testing

Data Managemen

dbt

Data Cleaning

Data Wrangling

ETL Pipelines

GCP

Snowflake

Feature Engineering

PostgreSQL

AI tools

LLMs

GitHub Copilot

Claude Code

Prompt Engineering

Data governance

Cursor

OpenAI

Codex

Business Analysis

Requirements Gathering

KPI Reporting

Process Documentation

Stakeholder

@ EXPERIENCE

AI Engineer

2024

• Designed and built a production AI system that ingests user stories and automatically generates structured test cases with acceptance criteria — eliminating hours of manual QA documentation per sprint cycle.

• Engineered the core NLP pipeline to parse free-form requirement text and produce structured, reproducible test case outputs, enabling QA teams to shift from writing to reviewing

Design engineer

2024 — NOW

Designed and built an admin panel for enterprise clients, scaling to support over 500 active users per instance.

Your description here

Software engineer at

2017 — 2020

Developed the user interface for a crypto payment gateway, ensuring compliance with global accessibility standards.

Your description here

@ EXPERIENCE

2024

AI Engineer

• Engineered the core NLP pipeline to parse free-form requirement text and produce structured, reproducible test case outputs, enabling QA teams to shift from writing to reviewing

2024 — NOW

Design engineer

Designed and built an admin panel for enterprise clients, scaling to support over 500 active users per instance.

Your description here

2017 — 2020

Software engineer at

Developed the user interface for a crypto payment gateway, ensuring compliance with global accessibility standards.

Your description here

@ SKILLS & TOOLS

Languages

& Tools

Languages & Tools

Python

SQl

Tableau

Power BI

Excel

Git

Jupyter

Google Colab

NoSQL

PySpark

Analytics & Statistics

EDA

Time-Series

ARIMA

A/B Testing

Regression

Statistical Modelling

Hypothesis Testing

Data

Management

Data Managemen

dbt

Data Cleaning

Data Wrangling

ETL Pipelines

GCP

Feature Engineering

PostgreSQL

Snowflake

Business Analysis

Requirements Gathering

KPI Reporting

Stakeholder Communication

Process Documentation

ai tools

AI tools

LLMs

Prompt Engineering

GitHub Copilot

Claude Code

Cursor

OpenAI

Codex

Data governance

Let's Work Together

Let's Work Together

Let's Work Together