I am Nitin Joseph. I am a self-taught ML, NLP (Natural Language programming) & python developer focusing on Marketing, Ecommerce, Retail & Analytics with deep skills in SEO, Content strategy & Martech. I can be contacted at

[email protected]

I have degrees in Quantitative disciplines and Liberal arts. I embrace STEAM mindset 🙌

One thing led to another when 5 years ago, I started having some fun with Python; as Feynman said if you want to master something, teach it; I volunteered to teach in an online boot camp for Afghan students. If any, lessons were mine to learn in dedication, application, and hope. Important sentiment where the world found itself in those times.

Measuring the impact of Digital media is often a straight shoot, but OOH marketing initiatives with barely any digital trail are painful ways of torturing data and yourself. Curious about a client's marketing initiative, I used Multiple Linear Regression seeking coefficients to explain various initiatives' impact. Minitab & SPSS (statistical software) are easy-to-use statistical tools, but I was soon faced with questions such as relaxing the assumptions, regularisation, and assumption of normality of data.

Then things converged, Statistics, coding & domain expertise. Borderline alchemy and a powerful alembic

Buzz around ML generally veers from one extreme to another, a magic formula; throw at it a bunch of data and hope it gives you what you want. The other reduces ML models to their outcomes - prediction & classification. Later a good summarization but misses the essence.

Accuracy of Models to predict and classify are no doubt important, but if you are like me, who values questions as much as answers, the journey of getting there is paved with a path of insights and out-of-sights.

In respecting privacy & client confidentiality, I have used public datasets.

1. Understand why customers churn & then predict who will.

We use Random Forest Classifier to grade factors that have the most impact on churn rate and use the model to predict customers likely to churn. Checkout the Project

2. Facebook Prophet to forecast Demand & quantify the affect of seasonality, weekends and holidays

Prophet’s advantage are — an analyst-in-the-loop forecaster with human-interpretable parameters, easily applied to business analytics use cases. It's however getting long in the tooth and more recent developments with models like Neural Prophet & Nixtla/statsforecast are more appealing by the day. Checkout the Project

3. Understand consumer preference with Conjoint analysis

Which involves analyzing user preference survey data, applying methods to determine how users weigh each attribute and using data to predict and map out what consumers like and dislike about a product. Checkout the Project

4. E-commerce Market Basket Analysis to make a product recommendation & calculate CLV (Customer Lifetime Value)

• Utilize two popular algorithms to create product recommendations and perform market basket analysis. I use collaborative filtering and a priori algorithm and implement both to generate product recommendations using first-party sales data. Checkout the Project

• Perform RFM (Recency, Frequency & Monetary Value) analysis and calculate CLV using BG|NBD Model. Checkout the Project

• Use unsupervised ML methods to segment customers into groups and find their key characteristics as it relates to business. This can then be used downstream in business and marketing process to better serve customers and improve outcomes for business and customers alike. Checkout the Project

5. Boost Campaign Conversion Rate with Decision Tree

Leverage Decision tree to improve to identify a subset of users with a higher likelihood of conversion works with digital and non-digital datasets. Checkout the Project

6. Marketing Modeling Mix with Facebook Robyn

This measurement model provides a big-picture view of all marketing efforts. This allows us to understand the incrementality of the entire cluster of marketing spend (Online, offline, brand, performance, email, town criers, whatever we spend money on). Oh, and we can then split by channel and get incremental Sales, CPS, google, TV, Meta, Print, and everything else.

7. Measuring Incrementality with Geo Lift?

Difficult problem to solve but thanks to Bayesian statistic this helps us get a grip on incrementality! Helps identify the results the company would NOT have received but for Marketing. Owned. Earned. Paid. Read why measuring incrementality is important Think with Google Checkout the Project

1. Content Similarity checker.

Most content similarity checkers available online use TextRank/Frequency checker or some levenshtein distance to measure the distance between two strings. In my project, I utilize Goose3 to scrape all URLs that are inputted as a list and provide similarity scores from BERT using Hugging Face transformer. BERT measures semantic relationship. Checkout the Project

2. Auto Summarizer with Google T5.

This is a project that I built to provide a robust summarizer for Meta descriptions using Google T5. Plugin in your Url's and it will output meta descriptions. Google T5 is a generative summarizer. End project use was to auto summarize Meta description and consumer reviews. Checkout the Project