A full ML pipeline to classify human emotional state from raw audio files recorded in workplace environments. Built with a focus on real-world noisy audio rather than clean benchmark recordings — making feature selection the critical engineering challenge.

Attribute	Details
Stack	Python · librosa · MFCC · scikit-learn · Signal Processing
Scale	End-to-end pipeline from raw audio ingestion to classified emotional output
Performance	Model selected via stratified cross-validation across multiple classifiers
Security	Handles real-environment noisy audio — not clean benchmark data
Impact	Applicable to workplace wellness monitoring and human-computer interaction
Repository

Extracted MFCC, pitch, energy, and spectral features using librosa to form a rich feature vector per audio sample. Trained and compared multiple classifiers before selecting the best performer using stratified cross-validation against a noisy, real-world audio dataset.

A content-based recommendation engine using K-Nearest Neighbors and cosine similarity on book metadata, with a feature space engineered beyond raw metadata for higher-quality retrieval.

Attribute	Details
Stack	Python · scikit-learn · KNN · Cosine Similarity · NLP · Pandas
Scale	Large book dataset vectorised for fast similarity-based retrieval
Performance	Improved accuracy through combined text-based and categorical feature representations
Security	Reproducible pipeline with version-controlled preprocessing steps
Impact	Personalised reading recommendations from lightweight, interpretable ML
Repository

Engineered a better feature space than raw metadata alone by combining text-based and categorical representations. Pre-processed and vectorised a large book dataset to enable fast and accurate similarity-based retrieval using KNN with cosine similarity.

Regression-based price prediction trained on a structured vehicle dataset, with an emphasis on handling messy, inconsistent real-world data and rigorous feature engineering before any modelling.

Attribute	Details
Stack	Python · scikit-learn · Regression · EDA · Pandas · NumPy
Scale	Full structured data pipeline from raw CSV to evaluated regression output
Performance	Feature selection, outlier handling, and normalisation applied to push model metrics
Security	Reproducible experiment configuration with clear preprocessing stages
Impact	Demonstrates end-to-end supervised regression on real-world tabular data
Repository

Performed thorough exploratory data analysis to identify the strongest predictors before any modelling. Applied feature selection, outlier handling, and normalisation — demonstrating that clean, well-engineered features matter more than model complexity on structured data.

A time-series forecasting pipeline for stock price prediction using regression-based models with temporal feature engineering, evaluated against standard financial forecasting metrics.

Attribute	Details
Stack	Python · scikit-learn · Time Series · Regression · Pandas
Scale	Multi-step lag feature pipeline with rolling statistics for temporal context
Performance	Evaluated using MAE, RMSE, and R² — hyperparameter-tuned with cross-validation
Security	Reproducible training with version-locked dependencies
Impact	Applied time-series ML to financial forecasting with interpretable regression models
Repository

Applied lag feature engineering and rolling statistics to provide the model with meaningful temporal context. Tuned hyperparameters with cross-validation and evaluated performance using MAE, RMSE, and R² metrics against held-out financial time series.

A creative text-generation system using NLP techniques for open-ended narrative creation, designed with a prompt-driven architecture that produces coherent, genre-aware story outputs.

Attribute	Details
Stack	Python · NLP · Text Generation · Generative AI
Scale	Open-ended narrative generation with prompt-controlled genre and style
Performance	Produces coherent, contextually consistent story outputs
Security	Prompt-driven architecture with controlled output boundaries
Impact	Demonstrates applied generative AI and creative NLP engineering
Repository

Designed a prompt-driven architecture that produces coherent, genre-aware story outputs using NLP techniques. The system handles open-ended narrative creation while maintaining thematic consistency across generated text.