End-to-end Data Science
Planning → data contracts → Airflow/dbt/Spark features → training/validation
(scikit-learn • XGBoost • PyTorch) → packaging & CI/CD → monitored deploy with MLflow.
Open to Data/ML roles & consulting
Hi, I'm Mayur Dalvi.
I’m a Data Scientist with 3.5+ years of professional experience building end-to-end data science products — from problem scoping and data contracts to feature pipelines, model training/validation, CI/CD, and monitored deployment. I’ve shipped forecasting, anomaly detection, and decision intelligence solutions used by business leaders.
I love tackling hard, ambiguous problems and turning them into usable, data-driven products. My goal is to leverage my skills to help companies empower users and confidently go after what’s next.
Python
SQL
Power BI (DAX)
Snowflake
Airflow
dbt
Spark
AWS
XGBoost
Prophet
Pandas
NumPy
Scikit-learn
TensorFlow
PyTorch
MLflow
Docker
Git
Linux / CLI
Tableau
Excel
Power Query
BigQuery
Redshift
Databricks
Azure
GCP
Looker Studio
Matplotlib
Seaborn
Plotly
Feature Engineering
EDA
Statistics
A/B Testing
Time Series
NLP
Computer Vision
Model Deployment
Monitoring
Data Modeling
ETL / ELT
What I like working on
Forecasting, Anomaly & Causal
Hierarchical/time-series (Prophet, ARIMA), change-point & anomaly detection,
uplift modeling and causal inference (DoWhy) with backtesting & guardrails.
Decision Intelligence & BI
Executive dashboards in Power BI with curated DAX, drill-through, RLS,
and a maintainable semantic layer aligned to business metrics.
LLM Systems & Agents
RAG architectures (chunking, embeddings, vector DBs), tool-use AI agents,
eval harnesses, prompt safety, and observability for production quality.
Work Experience
City & County of Denver — Data Analyst
- Automated audit workflows across 200+ locations with geospatial validations & policy guardrails.
- Shipped forecasting pack (Prophet / XGBoost) with backtesting & CV → accuracy +28%.
- Integrated Snowflake Document AI pipelines (streams/tasks) → key-field accuracy ~95%.
Python
SQL
Snowflake
Power BI (DAX)
Airflow
dbt
Deque Systems — Data Analyst
- Built learner analytics for 10k+ users; identified modules with +18% completion uplift.
- Delivered executive dashboards & retention cohorts → supported +15% renewals QoQ.
PythonSQLPower BICohort Analysis
ICR Inc — Data Scientist
- Ingested 10-year aviation + weather history; engineered features & trained XGBoost (ROC-AUC ≈ 0.80).
- Established reproducible pipelines with drift & anomaly monitoring for reliability.
Feature EngineeringModel EvalTime-series
NICE — Data Engineer
- Designed ETL across 10TB+ call data; parallelized Spark jobs → runtime 12h → 8h.
- Built Redshift analytics over billions of rows; storage & partitioning tuned for cost/perf.
SparkAirflowRedshiftData Modeling
Selected Projects
Credit Card Fraud Detection
Python • XGBoost • Scikit-learn • Power BI
Production-style pipeline for imbalanced classification: cost-sensitive training, threshold tuning, PR-AUC monitoring, and analyst workflow integration. GitHub →
Weather-Aware Flight Delay Modeling
Feature Engineering • Time-Series
Joined FAA/ASOS weather with flight histories; engineered lag features, evaluated with ROC-AUC across seasonal splits, and designed reproducible preprocessing. GitHub →
SageMaker Flight Price Prediction
AWS SageMaker • Jupyter • MLOps
End-to-end MLOps with experiment tracking, model registry, packaged inference, and deployment patterns for batch + near-real-time prediction. GitHub →
Movie Recommender System
Python • Similarity Search • TMDB
Content-based retrieval with vectorization + cosine similarity; fast prototyping, explainable results, and clean offline evaluation notebook. GitHub →
IPL Win Probability Predictor
Sports Analytics • Logistic Regression
Real-time feature updates drive calibrated win probability; showcases feature selection, calibration, and intuitive model outputs. GitHub →
Healthcare KPI Dashboard
Power BI • DAX • RLS
Executive KPIs for Patients, LOS, Cost/Stay with semantic model, curated DAX measures, and secure row-level security. GitHub →
A/B Testing — Digital Ads
Python • statsmodels
Statistically sound experimentation: Z-tests, lift & CIs, sanity checks, outlier handling, and reproducible workflow for marketing teams. GitHub →
Olympics Data Analysis
Jupyter • EDA • Viz
Historical Olympics EDA with tidy joins, feature discovery, and clear visual storytelling for non-technical stakeholders. GitHub →
Exoplanets — Data Visualization
Python • Matplotlib • Seaborn
Clean visualization of exoplanet catalogs covering scales, outliers, and correlations with disciplined chart design. GitHub →
Hollywood Market Synopsis Viz
Pandas • Plotting
Market KPIs for revenue, genres, ROI; robust CSV cleaning and reusable visualization helpers. GitHub →
Profitability Dashboard (Tableau)
Tableau • Data Modeling
Cohort and margin drill-downs; optimized extracts and pragmatic fact/dim semantics for speed. GitHub →
AI-Powered Automated Data Analysis
Python • LLM • Viz
LLM-assisted EDA & charting for CSVs with guardrails; prompts → insights + visuals for faster exploration. GitHub →
SourceSense — AI Agent
Agents • Tools • Retrieval
RAG agent with tool use; strong observability, error-handled boundaries, and clean tracing for debugging. GitHub →
AI Agents Chatbot (LangChain)
Streamlit • FastAPI • LangChain
Multi-model agent (Groq + OpenAI) with optional web search; clean API boundary and helpful UI. GitHub →
Flight Analytics App
Streamlit • MySQL • Plotly
Query & visualize airline mix, busiest airports, and daily frequencies with portable DB helpers. GitHub →
SmartTube Summarizer
Video • Summarization
End-to-end pipeline: transcription, chunking, ranked summaries, and a clean “summary UX”. GitHub →
Watch-Wise Tool
Scraping • Parsing
Streaming discovery helper using resilient scraping + parsing; lightweight CLI/app patterns. GitHub →
Click a category to filter. Images live in assets/projects/.
Let’s work together
Email: mayurdalvi.5@gmail.com
Location: Denver, CO (Open to relocate)