Resume/CV

Contact

Summary

Data Scientist with strong experience designing and deploying end-to-end machine learning systems, spanning data ingestion, statistical analysis, modeling, MLOps, and production deployment. Proven ability to work across time-series, signal processing, NLP, recommender systems, spatiotemporal forecasting, and database-backed analytics, with projects and internships translating theory into production-grade pipelines and applications.

Hands-on experience building reproducible ML workflows using DVC, MLflow/DagsHub, Optuna, Docker, and CI/CD, and deploying models via FastAPI, Streamlit, and AWS. Comfortable working from raw data and sensors through feature engineering, model evaluation, and scalable inference, with a strong foundation in statistical modeling, experimental rigor, and system design. Known for clear technical communication, disciplined engineering practices, and research-informed problem solving.

Work Experience

Machinery Monitoring Systems LLC — Remote, USA

Data Science / Machine Learning Intern
August 2025 – December 2025

Built a real-time end-to-end system for automated fault detection in rotating machinery using multivariate accelerometer time-series. Designed the full workflow covering data collection, preprocessing, feature engineering, modeling, real-time inference, and UI integration, with iterative experimentation using Agile practices.

Key contributions:

  • Built a full data collection pipeline for a 3-axis accelerometer connected to a rotor kit, capturing 1024-sample waveform frames at high sampling rates.
  • Developed a threaded serial-communication system (Python) to parse raw hex sensor streams, reconstruct waveforms, remove DC components, compute rFFT features, and store synchronized X–Y–Z triplets.
  • Collected and curated labeled datasets for normal, bearing fault, unbalance, misalignment, and looseness conditions; maintained clean, reproducible dataset organization.
  • Engineered time-domain and frequency-domain representations enabling dual-branch modeling (waveform + FFT).
  • Trained and evaluated advanced models including MiniROCKET + Ridge, 1D CNNs, InceptionTime, and fused waveform–FFT architectures.
  • Achieved ≈99% performance across accuracy / F1 / precision / recall / sensitivity / specificity on multi-class fault prediction (reported internally).
  • Built a Streamlit UI for real-time inference, waveform + FFT visualizations, confusion matrices, and model comparisons.
  • Conducted representation analysis using PCA and t-SNE to visualize fault separability in low-dimensional embeddings.
  • Followed software engineering best practices: modular codebase, OOP design, version control, reproducible pipelines, logging, and iterative experimentation.

Keywords: time-series classification, vibration analysis, signal processing, rFFT, streaming inference, MiniROCKET, CNN, InceptionTime, Streamlit, Python, MLOps, reproducibility, PCA, t-SNE, Agile

Indian Institute of Tropical Meteorology (IITM) — Pune, India

MS Student Data Science Intern / Research Intern
January 2022 – January 2023

Developed spatiotemporal deep learning models for forecasting environmental risk events (e.g., forest fires and stubble burning) using satellite imagery and atmospheric variables; supported operational decision-making and contributed to a peer-reviewed publication.

Highlights:

  • Designed and evaluated ConvLSTM-based spatiotemporal forecasting models over geospatial grids; achieved ≈0.8 temporal correlation for 1-day-ahead forecasts (reported in project work).
  • Built automated pipelines for scalable preprocessing, storage, and visualization; collaborated with domain scientists to translate real-world forecasting goals into implementable modeling tasks.
  • Contributed to publication linking predicted fire-burning patterns to PM2.5 emission insights.

Keywords: ConvLSTM, spatiotemporal forecasting, satellite imagery, environmental ML, geospatial grids, data pipelines, research collaboration

Projects

YouTube Comments Analyzer (Chrome Extension + ML Pipeline + FastAPI)

Tech: Python, scikit-learn, LightGBM, MLflow, Optuna, DVC, FastAPI, AWS (S3, ECR, EC2, CodeDeploy), Docker, Git, GitHub, GitHub Actions, JavaScript

  • Built an end-to-end sentiment analysis system for YouTube comments, including an ML pipeline, deployment automation, and a Chrome extension frontend.
  • Performed iterative EDA + preprocessing (missing values, duplicates, lemmatization, stopword handling with negation retention, n-grams, word clouds).
  • Built a baseline Random Forest + Bag-of-Words model and tracked experiments using MLflow on DagsHub.
  • Improved performance through staged experiments: BoW vs TF-IDF, unigram/bigram/trigram comparisons, feature-size tuning, imbalance handling (under-sampling, SMOTE variants).
  • Trained and tuned multiple models using Optuna, selecting LightGBM as the final model based on consistent cross-metric performance.
  • Developed a reproducible training pipeline using DVC stages (ingestion → preprocessing → feature engineering → training → evaluation → model registry).
  • Built a FastAPI backend to serve predictions and visualizations (sentiment pie chart, word cloud, monthly trend graph) to a Chrome extension UI.
  • Implemented CI/CD using GitHub Actions, including model tests (load/signature/performance thresholds), automated promotion to production, and Docker deployment to AWS.

Food Delivery Time Prediction (End-to-End Regression + MLOps)

Tech: Python, Pandas, NumPy, scikit-learn, Optuna, MLflow, DVC, FastAPI, Docker, AWS (S3, ECR, EC2, CodeDeploy)

  • Built an end-to-end delivery time prediction system (45,593 rows, 20 features) including EDA → modeling → MLOps → deployment-ready API using Python and scikit-learn.
  • Performed iterative data cleaning + profiling, fixing hidden missing values (string-encoded "NaN"), trimming whitespace noise, resolving datatype issues, and validating missingness patterns using missingno visualizations.
  • Engineered high-impact features such as city extraction from rider IDs, haversine distance + binned distance categories, and datetime features (weekday/weekend, time-of-day buckets, pickup time minutes) to improve predictive signal.
  • Conducted statistically driven EDA using Chi-square tests, ANOVA, and Jarque–Bera normality testing, identifying key drivers like traffic density, distance, festival effects, and time-of-day trends.
  • Developed reproducible preprocessing pipelines using ColumnTransformer + Pipeline, combining MinMax scaling, one-hot encoding, and ordinal encoding (traffic, distance bins), with Yeo–Johnson target transformation for improved regression behavior.
  • Established baseline regression models and evaluated trade-offs between dropping vs imputing missing values, demonstrating stronger results when dropping missing rows (MAE ≈ 4.73 min, R² ≈ 0.60 for linear regression).
  • Implemented advanced model experimentation with MLflow (DagsHub tracking) and Optuna Bayesian optimization, comparing Random Forest, Gradient Boosting, XGBoost, LightGBM, KNN, and SVM, selecting LightGBM as best-performing (MAE ≈ 3.05 min, R² ≈ 0.84).
  • Designed a full MLOps workflow with DVC pipelines (S3 remote) and production deployment plan using FastAPI + Swagger, Docker containerization, and scalable AWS architecture (EC2 auto-scaling + load balancer, API validation via Postman + stress testing).

Spotify Recommendation System (Hybrid Recommender + Scalable Processing)

Tech: Python, Streamlit, Pandas, NumPy, Dask, scikit-learn, SciPy, Matplotlib, Seaborn, DVC Pipelines, Git, GitHub, GitHub Actions, AWS (S3, ECR, EC2, CodeDeploy)

  • Built an end-to-end Spotify Hybrid Song Recommender System combining Content-Based Filtering (CBF) and Item-Based Collaborative Filtering (CF) to improve personalization, diversity, CTR, and reduce churn.
  • Performed extensive EDA on 50,683 Spotify tracks and 9.7M+ user–song interaction records, including missing value analysis, duplicate handling, and feature distribution insights.
  • Developed a content-based similarity engine using a rich feature pipeline with TF-IDF on tags, one-hot encoding for categorical metadata, frequency encoding, and scaling for audio features, followed by cosine similarity ranking.
  • Implemented item-based collaborative filtering using a massive user–item space (≈30K songs × ≈1M users) and solved the memory bottleneck by using Dask + sparse interaction matrices for scalable similarity computation.
  • Designed a weighted hybrid recommender that combines CBF and CF similarity scores with tunable weights, including fixes for index misalignment and score-scale mismatch using sorting + min-max normalization.
  • Built an interactive Streamlit web app allowing users to select recommendation mode (CBF / CF / Hybrid), choose number of recommendations, and dynamically adjust hybrid weighting through a slider.
  • Created a reproducible ML/data workflow using a DVC pipeline with modular stages (cleaning → preparation → transformation → recommendation), enabling consistent data + artifact versioning and fast iteration.
  • Deployed the application using production-style infrastructure: Dockerized the app, configured CI with GitHub Actions, stored artifacts via S3 + DVC, published images to ECR, and automated rollout using EC2 + CodeDeploy (blue/green) + Auto Scaling + Load Balancer.

NYC Taxi Demand Prediction (Time Series Forecasting + App)

Tech: Python, scikit-learn, LightGBM, MLflow, Optuna, DVC, FastAPI, AWS (S3, ECR, EC2, CodeDeploy), Docker, Git, GitHub, GitHub Actions, Streamlit

  • Built an end-to-end spatiotemporal taxi-demand forecasting system that predicts the next 15-minute pickup demand given a driver’s location (lat/long) and time, using NYC Yellow Taxi trips from January to March 2016.
  • Performed large-scale EDA with Dask to handle data that is too large for memory, identifying key demand drivers (especially time-of-day / day-of-week) and diagnosing extreme outliers in location/distance/fare fields.
  • Cleaned the dataset by removing erroneous outliers, then partitioned NYC into 30 data-driven regions using scaled lat/long + MiniBatchKMeans, selecting \(k\) by targeting ≈1–1.5 mile average distance to each region’s 8 nearest neighbors.
  • Converted raw trip logs into region-wise time series by resampling to 15-minute intervals and counting pickups per region, then applied EWMA smoothing (\(\alpha = 0.4\)) to stabilize short-term fluctuations.
  • Engineered forecasting features including 4 lag demands (\(t−1\) to \(t−4\)), day-of-week, region identifiers, and smoothed demand; used a time-aware split (train: Jan–Feb, test: Mar).
  • Trained and evaluated a linear regression baseline that achieved about 7.93% test MAPE (and ≈8.78% train MAPE), then ran Optuna-based model selection & hyperparameter tuning while tracking all runs in MLflow (hosted via DagsHub) and registering the best model.
  • Productionized the workflow as a reproducible DVC pipeline (data ingestion → cleaning → clustering → resampling/feature engineering → training → evaluation → registration), with artifacts versioned in S3 as the DVC remote.
  • Delivered a Streamlit app that visualizes predicted demand on a region map (optionally current + 8 neighbors only), and implemented CI tests (pytest) + CI/CD to containerize with Docker, push to ECR, and deploy via CodeDeploy to an EC2 Auto Scaling Group behind a load balancer.

Olist E-Commerce Database (PostgreSQL → SQLite + Streamlit App)

Tech: SQL, PostgreSQL, SQLite, ER modeling, BCNF, Functional Dependencies, Joins, Aggregations, Subqueries, Window Functions, Indexing, Query Optimization, EXPLAIN, Streamlit

  • Built an end-to-end relational database application for the Olist Brazilian e-commerce dataset (≈100K orders, 2016–2018), supporting customers, sellers, products, orders, payments, reviews, and geolocation analytics.
  • Designed and implemented a normalized PostgreSQL schema (RAW) with primary keys, foreign keys, and relationship constraints reflecting marketplace entities and one-to-many relationships (customers → orders, orders → items/payments/reviews, sellers → items).
  • Performed global BCNF verification by identifying candidate keys + functional dependencies across all relations; introduced surrogate keys (geolocation_id, review_record_id) to resolve duplicate-record key issues and enforce BCNF at the schema level.
  • Developed and validated core CRUD + analytics SQL: insert/update/delete workflows plus multi-table SELECTs for top revenue products, seller rating summaries, high-spending customers, and low-rated delivered orders.
  • Authored and executed advanced analytical queries using subqueries and window functions (e.g., PERCENT_RANK) to rank sellers within states and compute revenue/behavior metrics.
  • Conducted query performance analysis using EXPLAIN outputs, identifying expensive joins/group-bys and proposing improvements via indexing and better query structure for large-table scans.
  • Implemented indexing strategies (B-Tree; and discussed Hash indexes in PostgreSQL) and compared before/after runtime on heavy join queries (measured execution time improvements).
  • Migrated the system from PostgreSQL → SQLite using Python scripts (create_db.py, index_db.py) and deployed an interactive Streamlit web app enabling users to run custom SQL queries against the dataset.

Technical Skills

Programming & Software Engineering

  • Python (advanced): modular codebases, OOP design, threading & multiprocessing, structured logging
  • R: statistical modeling, regression, time series analysis, forecasting
  • SQL: complex joins, subqueries, window functions, query optimization (EXPLAIN)
  • MATLAB: numerical computing and signal processing (academic use)
  • Scripting & automation; reproducible research writing (LaTeX, Quarto, Markdown)

Data Science & Statistical Modeling

  • Exploratory Data Analysis (EDA) at scale (Pandas, Dask)
  • Feature engineering for tabular, time-series, text, and geospatial data
  • Regression, classification, and predictive modeling
  • Model evaluation: cross-validation, bias–variance analysis, calibration, diagnostics
  • Statistical testing: ANOVA, chi-squared tests, correlation analysis
  • Dimensionality reduction & representation analysis (PCA, t-SNE)

Time Series, Signal Processing & Spatiotemporal ML

  • Time-series forecasting and classification
  • Signal preprocessing: DC removal, normalization, smoothing (EWMA)
  • Frequency-domain analysis: rFFT, spectral feature extraction
  • Waveform-based modeling (raw signals + engineered spectra)
  • Spatiotemporal modeling on geospatial grids
  • Temporal resampling, lag feature construction, rolling statistics

Machine Learning & Deep Learning

  • Classical ML: Linear & Ridge Regression, Logistic Regression, Random Forests, LightGBM
  • Time-series ML: MiniROCKET, feature-based classifiers
  • Deep Learning:
    • 1D CNNs for sequence modeling
    • InceptionTime architectures
    • ConvLSTM for spatiotemporal forecasting
  • Model comparison across architectures and feature representations
  • Hyperparameter optimization using Optuna

Natural Language Processing (NLP)

  • End-to-end sentiment analysis pipelines
  • Text preprocessing: normalization, lemmatization, stopword handling (with negation retention)
  • Feature extraction: Bag-of-Words, TF-IDF, n-grams (uni/bi/tri-grams)
  • Class imbalance handling (under-sampling, SMOTE variants)
  • Model selection and robustness evaluation

Recommender Systems

  • Content-Based Filtering using TF-IDF, one-hot & frequency encoding, scaled audio features
  • Item-Based Collaborative Filtering
  • Hybrid recommender design with weighted similarity fusion
  • Large-scale similarity computation using Dask and sparse matrices
  • Recommendation ranking, normalization, and evaluation strategies

Data Engineering & Pipelines

  • End-to-end ML and data pipelines
  • Dataset versioning and lineage tracking with DVC
  • Pipeline orchestration using DVC stages
  • Scalable data ingestion, cleaning, transformation, and artifact management
  • Schema-aware preprocessing and data validation

MLOps & Experimentation

  • Experiment tracking with MLflow (hosted via DagsHub)
  • Metric logging, artifact tracking, and model registry workflows
  • Reproducible experimentation and ablation studies
  • Model testing: loadability, signature checks, performance thresholds
  • CI-based validation prior to production promotion

Deployment, APIs & Applications

  • FastAPI for model serving and analytics APIs
  • Streamlit for interactive ML applications and dashboards
  • Dockerized ML services and applications
  • REST-based inference services with visualization endpoints
  • Backend integration for browser-based clients (Chrome Extension)

Cloud & DevOps

  • AWS: S3 (artifact storage), ECR, EC2, CodeDeploy
  • Containerized deployment workflows
  • CI/CD using GitHub Actions
  • Blue/green deployment strategies
  • Auto-scaling and load-balanced architectures (applied)

Databases & Analytics Engineering

  • Relational schema design and ER modeling
  • Database normalization up to BCNF
  • Functional dependency analysis
  • PostgreSQL, SQLite
  • Indexing strategies (B-Tree; PostgreSQL Hash index discussion)
  • Analytical SQL for business and performance insights

Visualization & Reporting

  • Matplotlib, Seaborn
  • Time-series visualizations
  • Confusion matrices, ROC / PR analysis
  • Streamlit dashboards for real-time inference and analytics
  • Technical reporting for academic and production contexts

Reproducibility, Tooling & Collaboration

  • Git-based version control
  • Reproducible ML workflows and pipelines
  • Modular, testable codebases
  • Agile experimentation and iterative development
  • Cross-functional collaboration with domain experts

Education

University at Buffalo, SUNY

MS in Engineering Science (Data Science) — GPA: 4.0/4.0
August 2024 – December 2025

Courses

  • Introduction to Numerical Mathematics for Computing and Data Scientists
  • Introduction to Probability Theory for Data Scientists
  • Programming and Database Fundamentals for Data Scientists
  • Statistical Learning and Data Mining - I
  • Statistical Learning and Data Mining - II
  • Predictive Analytics
  • Data Intensive Computing
  • Introduction to Machine Learning
  • Data Models Query Language
  • Experiential Projects in Artificial Intelligence and Data Science

Indian Statistical Institute (ISI) Kolkata

Post-Graduate Diploma in Applied Statistics (Data Analytics) — GPA: 9/10
September 2023 – August 2024

Courses

  • Basic Statistics
  • Basic Probability
  • Statistical Methods
  • Survey Sampling
  • Introduction to Official Statistical System
  • Statistics and Economy
  • Introduction to R and Python
  • Multiple Regression
  • Advanced Regression
  • Time Series Analysis and Forecasting
  • Multivariate Statistics
  • Statistical Machine Learning

Indian Institute of Science Education and Research (IISER) Pune

Integrated BS–MS Dual Degree (Science)
August 2017 – January 2023

Publications