We help youretain customerspredict demandtrack every rupeeautomate reportswith the data you already keep.

Customer Intelligence and Churn Platform

A full customer-analytics system for a subscription business. Cleaned data sits in MySQL, 37 SQL business questions answer the common management questions, and three models cover churn, customer segments and lifetime value, all shown in a 7-page Streamlit dashboard. An AI copilot answers plain-English questions with checked, read-only SQL, summarises customer feedback and suggests retention actions.

In a simulated retention campaign with stated costs, contacting customers by model ranking beat contacting everyone by about 54%. The churn signal is modest (ROC-AUC 0.62).

Restaurant Customer Intelligence

Takes messy restaurant and order data and turns it into a dashboard a manager can read. Twelve raw datasets were cleaned in MySQL, then explored with 26 SQL queries (CTEs and window functions), two reusable views and a 30-chart analysis.

Delivered a 6-page Power BI dashboard and a churn model (ROC-AUC 0.76) that flags customers likely to leave.

Patient Readmission Analysis

Studies which patients return to hospital and why. We analysed over 100,000 diabetic patient records from 130 US hospitals using Python and advanced SQL, then built an interactive dashboard with KPIs and risk groups.

Found readmission above 48% in patients aged 70 to 90, and 68.58% in the Very High Risk group.

Customer Churn Prediction (Banking)

Predicts which bank customers are likely to leave so a team can contact them first. The model was tuned for recall because missing a leaving customer costs more than a false alarm. It runs behind a REST API with a simple web screen for non-technical users.

About 91% recall on the test data.

Loan Default Risk Scoring

Scores how likely a loan applicant is to default, so an underwriter can see risk at a glance. Customer, credit-bureau, loan and transaction data was loaded into a MySQL database, analysed with 10 SQL queries, and used to train and compare four models. The output is a 0 to 100 risk score (Low, Medium, High) in a Streamlit dashboard with filters and a high-risk watchlist.

In testing, loans in the High group defaulted about 46% of the time, against about 10% in the Low group (ROC-AUC 0.72 on held-out data).

E-commerce Purchase Intent

Predicts whether an online shopping session will end in a purchase, and explains every prediction with SHAP and LIME so the result can be trusted and acted on. It is built as a full MLOps pipeline: gradient-boosting models, experiment tracking in MLflow, a real-time FastAPI prediction service, an interactive Streamlit dashboard, Docker packaging, and automated tests and checks on GitHub.

NHS Mental Health Data Analytics

Analytics work on large UK health datasets. We built an ETL pipeline that replaced a manual monthly upload, grouped patients with K-Means clustering, flagged unusual prescribing with Isolation Forest, and tested which risk factors were statistically significant.

Pipeline handles 770,000+ rows. Isolation Forest flagged 318 of 6,360 records. The clustering is exploratory (silhouette 0.20).

AI Customer Support Agent

Reads incoming support emails, works out the intent, drafts a reply from a company knowledge base, and passes low-confidence or sensitive cases to a person. It includes protections against prompt-injection, retry handling and a 36-query automated test suite to check answer quality.

Text Emotion Classification

Trains and compares four neural network designs (RNN, GRU, Bidirectional GRU and LSTM) to recognise emotion in text, with a web front end to try it out.

Sentiment and Intent Modules

Language-understanding components built for an AI-based mental health assistance system, covering data preparation, feature engineering and deployment as part of a cross-functional team.

AI Content Automation Pipeline

A four-stage workflow that finds trending topics, writes prompts, drafts blog content with an LLM and logs everything to a spreadsheet, with no manual steps in between. It runs in Docker.

MindCare Wellness AI Agent

A conversational agent for a wellness clinic. It answers common patient questions from a knowledge base, captures appointment enquiries straight into Salesforce, scores each as Hot, Warm or Cold, and logs the conversation as a linked task. A live demo is available on request.