Manjit Singh

Data Science Portfolio

---

Education

Executive Post Graduate Program in Data Science & AI
IIIT Bangalore
Oct ‘23 – Jan ‘25

Bachelor of Commerce (B.Com)
Gauhati University
Jul ‘20 – Jun ‘23

---

Professional Experience

Associate - Data Analyst
Samhita | Remote | Nov 2024 – Present

At Samhita, I support data-driven decision-making for financial inclusion projects by managing datasets, uncovering insights, and automating processes across multiple programs. I work closely with program, tech, and MEL teams to ensure data flows efficiently from collection to actionable insight.

• Key Responsibilities:

🧹 Data Management & Reporting:

• Maintain and update datasets across Access to Finance initiatives including Pre-Credit Score (PCS), Returnable Grants, and Formal Loans.
• Consolidate repayment data from various sources and support data cleaning, validation, and standardization.

📊 Dashboarding & Visualization:

• Assist in building and maintaining dashboards for internal monitoring and external reporting.
• Support real-time updates and enhancements for evolving project needs.

🔍 Data Analysis & Insight Generation:

• Conduct exploratory data analysis (EDA) to identify trends and anomalies.
• Extract, transform, and analyze data from SQL databases to inform strategic decisions.

⚙️ Process Automation & GenAI Adoption:

• Develop automation scripts using Python, SQL, and VBA to streamline recurring data workflows.
• Contribute to identifying and piloting GenAI use cases to enhance team productivity.

🤝 Cross-Team Collaboration:

• Coordinate with cross-functional teams to respond to data requests and support grant reports, donor updates, and tool development.
• Assist in the design of data collection formats and tracking templates for improved monitoring.

---

Internships

Business Analyst Intern
Quest Global Technologies Ltd | Remote | Nov 2023 – May 2024

• Led the automation of reporting workflows using Excel, SQL, and visualization tools like PowerBI, enabling real-time KPI tracking and reducing manual effort by 10%.
• Designed and analyzed A/B testing strategies to evaluate marketing effectiveness, resulting in a 5% boost in conversion rates and improved ROI.
• Applied data cleaning, segmentation, and exploratory analysis to uncover business patterns and improve customer targeting.
• Collaborated with product and sales teams to deliver data-driven insights, enhancing operational decision-making and customer engagement.
• Created reproducible reports, dashboards, and documentation to support strategic planning and ensure data transparency across teams.

---

Key Projects

🛍️ Sentiment-Based Product Recommendation System

Tech Stack: Python, Flask, Scikit-learn, Pandas, NLTK, HTML/CSS

Built an intelligent recommendation engine that combines collaborative filtering with sentiment analysis to generate more accurate and personalized product suggestions.

• Problem Addressed: Traditional recommender systems often overlook the sentiment expressed in user reviews, which can lead to irrelevant recommendations.
• Solution Overview:
  • Performed text preprocessing using NLTK and extracted features using TF-IDF vectorization.
  • Applied Logistic Regression to classify the sentiment (positive/negative) of product reviews.
  • Enhanced collaborative filtering by integrating sentiment scores to refine recommendation relevance.
  • Developed a Flask-based web app to interactively serve recommendations based on user input.

✅ Impact: Improved recommendation precision by aligning suggestions with user sentiment, resulting in a more engaging and tailored user experience.

📄 Helmate AI: Retrieval-Augmented Generation (RAG) System

Tech Stack: Python, LangChain, GROQ AI, SBERT, ChromaDB

Developed a cost-efficient, open-source RAG pipeline to accurately respond to user queries based on insurance policy documents, enhancing the accessibility and understanding of complex policy information.

• Problem Addressed: Navigating large volumes of insurance documentation can be tedious for users looking for specific policy-related answers.
• Solution Overview:
  • Implemented semantic search using SBERT embeddings to retrieve contextually relevant chunks.
  • Used ChromaDB as the vector store for efficient document retrieval.
  • Integrated LangChain to orchestrate the pipeline and GROQ AI as the LLM for generating natural, context-aware responses.
  • Optimized the workflow to ensure fast, accurate, and scalable information retrieval across diverse policy documents.

✅ Impact: Delivered a scalable and interpretable RAG system with improved query response accuracy—built entirely on open-source tools for maximum accessibility and zero API cost.

🧾 Automatic Ticket Classification System

Tech Stack: Python, Scikit-learn, NLTK, Pandas

Built an intelligent automated ticket classification system to streamline customer support by accurately routing tickets to the relevant departments.

• Problem Addressed: Manual triaging of support tickets was time-consuming and error-prone, often delaying response times and affecting customer satisfaction.
• Solution Overview:
  • Applied text preprocessing and NLP techniques (tokenization, stopword removal, lemmatization) using NLTK.
  • Performed topic modeling with Non-negative Matrix Factorization (NMF) to uncover hidden structures in unstructured ticket texts.
  • Built and evaluated multiple classifiers using Scikit-learn for efficient ticket label prediction based on content.
  • Developed a reproducible Jupyter Notebook pipeline, allowing for future scaling and integration.

✅ Impact: Automated classification significantly reduced manual workload, improved response speed, and laid a foundation for intelligent support systems.

🧩 Customer Segmentation using Clustering

Tech Stack: Python, Scikit-learn, Pandas, Seaborn, Matplotlib

Implemented a robust unsupervised learning pipeline to segment customers based on purchasing behavior for targeted marketing and business strategy enhancement.

• Objective: Helped businesses uncover meaningful customer groups by analyzing transaction data and identifying key behavioral patterns.
• Approach:
  • Performed data cleaning, feature engineering, and RFM analysis (Recency, Frequency, Monetary) to derive meaningful metrics.
  • Applied K-Means and Hierarchical Clustering to segment customers and validate clusters using silhouette scores and dendrograms.
  • Visualized insights through heatmaps, cluster plots, and bar charts to help decision-makers understand segment characteristics.
  • Delivered actionable recommendations for tailored engagement, retention strategies, and personalized offers.

✅ Impact: Enabled data-driven customer targeting and improved marketing ROI through insight-backed segmentation.

🎬 RSVP Movies Analysis – SQL-Based Data Exploration

Tech Stack: MySQL, SQL Joins, Aggregations, Subqueries

Performed a comprehensive SQL analysis on the RSVP Movies dataset to derive actionable insights into movie performance, viewer behavior, and content trends.

• Objective: Analyze a real-world movie dataset to uncover trends related to movie views, genres, languages, platforms, and user engagement.
• Approach:
  • Wrote optimized SQL queries to explore viewer demographics, most-watched genres/languages, and regional popularity.
  • Used joins, window functions, subqueries, and aggregations to extract multi-dimensional insights across different entity tables.
  • Investigated monthly trends in user ratings and view counts to identify periods of peak engagement.
  • Segmented data by platform and genre to assist content curation decisions for OTT platforms.
  • Delivered conclusions in a structured manner for use in dashboards or further analysis.

✅ Impact: Enabled data-driven decisions for movie curation, marketing strategies, and platform-specific content optimization through insightful SQL reporting.

📉 Telecom Churn Prediction

Tech Stack: Python, Scikit-learn, Pandas, Seaborn, Matplotlib

Developed a predictive machine learning model to identify telecom customers likely to churn, enabling proactive retention strategies.

• Objective: Empowered telecom providers to reduce customer churn by predicting at-risk customers using historical usage and service data.
• Approach:
  • Conducted extensive EDA to uncover trends and patterns impacting churn (e.g., service issues, contract type, tenure).
  • Handled missing values, encoded categorical features, and applied feature selection techniques to improve model performance.
  • Trained and compared multiple classification models (Logistic Regression, Decision Trees, Random Forest, XGBoost), tuning hyperparameters for optimal results.
  • Evaluated performance using confusion matrix, ROC-AUC, precision, and recall to ensure model reliability and interpretability.
  • Visualized key insights with bar plots, correlation heatmaps, and model comparison charts.

✅ Impact: Equipped stakeholders with a reliable churn prediction system, enabling targeted interventions and improved customer retention.

🎯 Lead Scoring Model

Tech Stack: Python, Scikit-learn, Pandas, Seaborn, Matplotlib

Built a machine learning model to identify and prioritize marketing leads with the highest potential to convert, helping businesses optimize their sales funnel.

• Objective: Streamlined the lead qualification process by scoring leads based on behavioral and demographic attributes to maximize conversion rates.
• Approach:
  • Performed thorough data cleaning, feature engineering, and outlier treatment to ensure data quality.
  • Conducted EDA to uncover key patterns, such as lead source, total time spent, and page activity influencing conversion likelihood.
  • Applied Logistic Regression and Random Forest classifiers with grid search-based hyperparameter tuning.
  • Evaluated model using ROC-AUC, F1-score, and confusion matrix to balance precision and recall for business decisions.
  • Presented key takeaways through visual dashboards and feature importance charts for actionable stakeholder insights.

✅ Impact: Enabled targeted sales efforts by automating lead prioritization, thereby reducing time-to-conversion and improving ROI.

🚗 Vehicle Dataset - Advanced EDA & Insights

Tech Stack: Python, Pandas, Seaborn, Matplotlib, Plotly

Conducted an in-depth Exploratory Data Analysis (EDA) on a vehicle dataset to uncover trends, patterns, and actionable insights in the automotive space.

• Objective: Explored and visualized key factors influencing vehicle pricing, fuel efficiency, and design to support data-driven decisions for buyers and sellers.
• Approach:
  • Handled missing values, duplicates, and formatted features for consistency.
  • Applied univariate, bivariate, and multivariate analysis to understand the influence of attributes like engine size, brand, fuel type, and body style.
  • Leveraged correlation matrices, pair plots, violin plots, and box plots to visually interpret data relationships.
  • Used interactive Plotly dashboards for deeper insights and stakeholder-friendly visualizations.
  • Summarized insights with clear narratives and business implications after each section to make the notebook self-explanatory.

✅ Impact: Revealed strong correlations between engine size, brand, and price; highlighted the impact of fuel type on efficiency; and created a foundation for future predictive modeling.

🚴‍♂️ Bike Sharing Demand Prediction – Linear Regression Model

Tech Stack: Python, Pandas, Seaborn, Matplotlib, Scikit-learn

Built a linear regression model to predict bike rental demand based on historical usage patterns and environmental conditions.

• Objective: Analyze factors influencing daily bike rentals and develop a predictive model to estimate rental counts for a bike-sharing service.
• Approach:
  • Conducted EDA to understand seasonal trends, weather impact, and user behavior across weekdays vs. weekends.
  • Applied feature engineering to extract relevant time-based and categorical features (e.g., working day, holiday, temperature bins).
  • Built and evaluated a multiple linear regression model, interpreting coefficients to assess variable influence.
  • Used R² score, RMSE, and residual plots to assess model accuracy and refine performance.
  • Visualized relationships with scatter plots and heatmaps for better insight into dependencies.

✅ Impact: Provided an interpretable model that helps stakeholders anticipate rental demand and make informed operational decisions such as fleet planning and staff allocation.

💳 Credit Data Analysis – Deep Dive into Borrower Behavior

Tech Stack: Python, Pandas, Seaborn, Matplotlib, Plotly

Performed a comprehensive Exploratory Data Analysis (EDA) on a credit dataset to understand patterns related to credit risk, borrower profiles, and loan characteristics.

• Objective: Investigated how factors like age, income, loan purpose, credit history, and employment status impact loan approval and creditworthiness.
• Approach:
  • Conducted data cleaning, feature formatting, and null value handling for analysis readiness.
  • Performed univariate and bivariate analysis to explore distributions and identify key differentiators among defaulters and non-defaulters.
  • Utilized heatmaps, bar plots, histograms, and interactive visualizations to reveal correlations and highlight risk indicators.
  • Analyzed demographic and financial features in relation to loan status and provided clear, intuitive visual storytelling.
  • Integrated domain-relevant business interpretations to derive real-world insights from patterns in the data.

✅ Impact: Enabled a clearer understanding of high-risk segments, borrower behavior, and key attributes associated with loan defaults—laying groundwork for future predictive modeling.