Romika Solanki
Analyst
Python
SQL
Data Pipeline & Feature Engineering: 5/10
Experimentation & Evaluation: 4/10
Active 4 hours ago
+91 (9873) 515560
Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
A middle-level ML practitioner focusing on applied marketing-mix modeling and business analytics. The strongest proven skill is building end-to-end MMM pipelines including media-to-week alignment and adstock transformation as implemented in capstone_project.ipynb (adstock_transform and run_mmm_models_for_target). Public work lacks executed, reproducible artifacts and production readiness including pinned environments, time-series cross-validation, and robust uncertainty quantification.
Technical skills
Python
SQL
Python
SQLAlchemy
Databases
FAISS
SQLite
AI/ML
Embeddings
LangChain
NumPy
Pandas
RAG
Scikit-learn
Semantic Search
Semantic Search
AI/ML
Jupyter Notebook
Classic ML
DevOps
GitHub
Vector
Analytics
A/B Testing
Plotly
Power BI
Tableau
Matplotlib
ETL/ELT
Analytics
Timeline
Liverpool John Moores University
Master's Degree •
Data Science & AI
Middle AI/ML Engineer
Confidence: Medium Data-centric
Data-centric analytics practitioner (middle level) focused on e-commerce marketing mix modelling and feature engineering. The strongest proven skill is data pipeline and feature work, exemplified by the monthly-to-weekly media allocation and adstock feature construction in ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb. There is limited evidence of productionization, reproducible experiment tracking, test coverage or advanced ML/serving engineering in public code.
Model Architecture & Training
3/10
How well models are designed and trained
Basic, correctly applied predictive modeling using classical ML (linear regression) with reasonable transformations and a reusable modeling function, but no advanced architectures or training rigour.
Evidence
ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb: run_mmm_models_for_target (implements basic/log/multiplicative LinearRegression pipelines and metric reporting)
ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb: basic_model / log_model / mult_model training and metric calculation (r2, RMSE, MAE)
ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb: adstock_transform and adstocked media feature creation used as model inputs
Data Pipeline & Feature Engineering
5/10
How data is prepared for models
Solid data engineering and feature work including time-window alignment, media-month-to-week allocation, adstocking, holiday/payday flags, lags and rolling features.
Evidence
ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb: monthly-to-weekly media allocation loop building weekly_media_df and validation_df (reconstruction check)
ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb: feature engineering blocks (is_payday, is_holiday, temp_category, is_rainy, heavy_rain, extreme_temp)
ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb: lag and rolling features (gmv_lag1, gmv_roll_mean) and data cleaning (dtype coercion, dropna)
Experimentation & Evaluation
4/10
How results are measured and tested
Reasonable experimental comparisons and validation (train/test split, multiple model forms, metric reporting) plus a validation step for media allocation, but no reproducible experiment tracking or cross-validation and limited robustness analysis.
Evidence
ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb: train/test split usage and computation of r2, RMSE, MAE for each model variant
ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb: run_mmm_models_for_target returns a results DataFrame comparing model forms
ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb: validation_df comparing reconstructed_monthly to original_monthly to validate allocation logic
MLOps & Deployment
How models are shipped to production
Not evidenced in public code
Computational Efficiency
How efficiently computing resources are used
Not evidenced in public code
Research Depth & Innovation
Depth of research and new ideas
Not evidenced in public code
Expertise
E-commerce & Retail AI• Middle
Industries
Commerce• Middle
Data & Analytics• Middle
Sales & Marketing• Middle
Technologies
Classic ML
Python
SQL
LangChain
FAISS
SQLAlchemy
Embeddings
SQLite
RAG
Vector
Semantic Search
GitHub
Semantic Search
Recommendations
- Harden the workflow into reproducible scripts or a pipeline (Airflow/Prefect or simple CLI): parameterize paths, add unit tests for data transforms, and add a requirements file or environment spec.
- Introduce cross-validation and regularization (Ridge/Lasso/ElasticNet) and report stability (CI for coefficients) to address multicollinearity and small-sample risk.
- Add lightweight experiment tracking (MLflow or Weights & Biases) and a serialization step for best models to enable repeatable evaluation and later deployment.
- Refactor the monthly-to-weekly allocation and adstock logic into small, well-tested functions or a feature-engineering module so it can be reused and maintained.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Data Scientist
Confidence: High ML Practitioner
A middle-level ML practitioner focusing on applied marketing-mix modeling and business analytics. The strongest proven skill is building end-to-end MMM pipelines including media-to-week alignment and adstock transformation as implemented in capstone_project.ipynb (adstock_transform and run_mmm_models_for_target). Public work lacks executed, reproducible artifacts and production readiness including pinned environments, time-series cross-validation, and robust uncertainty quantification.
Statistical Rigor
3/10
Correct use of statistics
Basic validation metrics and sensible transforms are present but formal statistical checks, uncertainty quantification, hypothesis testing, and time-aware CV are missing.
Evidence
Data Wrangling & Cleaning
6/10
Preparing and cleaning data
Solid data wrangling and alignment work including type coercion, date parsing, weekly aggregation, and proportional monthly-to-weekly media allocation with a reconstruction validation step.
Evidence
Exploratory Analysis & Visualization
5/10
Exploring and visualizing data
Reasonable EDA with plots and written interpretations after findings, though largely descriptive and limited to conventional charts without deeper causal probing.
Predictive Modeling
4/10
Building models that predict
Multiple model formulations (linear, log, multiplicative) and adstock feature engineering are implemented, but modeling lacks CV, regularization, uncertainty estimates, and deeper error analysis.
Business Insight & Impact
6/10
Turning analysis into business value
Clear business framing and honest interpretation of negative out-of-sample performance with actionable recommendations tied to data limitations and multicollinearity concerns.
Reproducibility & Notebook Hygiene
2/10
Clean, repeatable analysis
Some reusable functions are defined but the notebook was not executed, uses hard-coded Google Drive paths, and lacks environment specification, seeding, data versioning, or reproducible run scripts.
Evidence
romikasolanki/ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb:# NOTEBOOK META (exec=none) and drive.mount('/content/gdrive') usage
romikasolanki/ECOMMERCE-MARKETING-MIX-MODEL/capstone_project.ipynb:adstock_transform and run_mmm_models_for_target defined inline rather than as reproducible modules
Industries
Commerce• Middle
Transportation & Logistics• Middle
Technologies
AI/ML
Analytics
Jupyter Notebook
Scikit-learn
Matplotlib
Pandas
NumPy
ETL/ELT
Recommendations
- Execute the notebook and capture outputs, add an environment spec (requirements.txt or pipenv/conda) and set deterministic seeds for reproducibility.
- Adopt time-aware validation (time-series CV) and regularized models (Ridge/Lasso) or Bayesian intervals to address overfitting and quantify uncertainty.
- Refactor key steps (media allocation, adstock, model training) into small, tested Python modules or scripts and add simple unit tests for the media allocation logic and adstock function.
- Add data versioning or sample fixtures and avoid hard-coded Google Drive paths by parameterizing data locations or using relative paths or a config layer.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: