I build the model,
and the page
it lives on.

Nine projects, one format.

Problem → method → evaluation → decision
01Optimisation

Route Optimiser

Multi-day visit planning: k-means on real travel-time matrices, nearest-neighbour ordering, opening-hours constraints.

−33% travel time vs planning in listed order
02Causal

Marketing Attribution Lab

50k users with known ground truth; naive attribution vs a randomised experiment.

Best on the dashboard, 2nd-worst in the experiment
03Forecasting

Demand Forecasting

Prophet, SARIMA and GBM on a 60-day holdout, then reorder points at 95% service level.

−82% stockout days · −84% lost sales
04Analytics

Mograh Analytics

GA4 plan, SQL reporting layer, Looker Studio, RFM segmentation and LTV for a womenswear brand.

914 of 920 buyers bought exactly once
05Healthcare ML

ECG Arrhythmia Classifier

Patient-level splits on MIT-BIH, imbalance-aware evaluation, MLflow tracking.

92.3% accuracy, 0% recall on the beats that matter
06AI systems

RAG Document Search

Chunking, embeddings, vector store and grounded generation, with a Streamlit front.

10/10 sourced, 10/10 grounded on a test set
07Audio ML

Speech Emotion Recognition

Transfer-learned CNN on speech clips; error analysis on where emotions blur.

70%+ across ten classes, 100% on gender
08Product

The Orchid Project

Music, journaling and self check-ins for student wellbeing, an initiative I founded.

Piloted with students, positive feedback
09Client work

Vinod Rice Mill

Role-based production management for a real manufacturer.

The first real client
About

A data and ML person who can also design, so the analysis and the thing people use are made by the same hands.

CertifiedGA4 · Google Ads · CITI
PublishedIJARIIT, 2023
CurrentlyAI & Data Science Intern, Broadplace · CS at KCL, 2027