AI systems, data engineering, research, and frontend — things I've built, shipped, and learned from.
A production AI agent handling customer queries and qualifying leads in real time. Combines natural language understanding with structured lead capture — turning conversations into business outcomes. Deployed on Railway.
Architected the retrieval infrastructure for an internal Q&A tool at Accenture (NatWest Group). Vector Databases serve unstructured banking documents to an LLM — deployed in a live banking environment.
Full-stack MVP for AI-powered marketing content generation. TypeScript frontend with a Python backend — generates campaign copy, ad variations, and content briefs using LLMs. Separate frontend/backend architecture for scalable deployment.
End-to-end Python pipeline solving the 'black box' problem in clustering. Automates dimensionality reduction (UMAP, t-SNE, PCA) and generates human-readable cluster interpretations via SHAP values and decision rules. MSc thesis — University of Greenwich.
Predictive quality-control model tackling two challenges: forecasting metal part lifespan from manufacturing parameters, and classifying components as defective or non-defective. Enables manufacturers to flag faulty parts before they reach production.
Bachelor's project that extracts student result data from Delhi University-issued PDFs, processes and cleans it, then visualises pass/fail and score distributions as pie charts. Automated pipeline from PDF intake to visual output using Python.
End-to-end data platform for the ShopSphere e-commerce domain. Ingests raw order data from MinIO (S3-compatible) into PostgreSQL, then builds analytics-ready staging, fact, and KPI models with dbt. Dockerised infrastructure with Python ingestion jobs and Airflow orchestration scaffolding.
Interactive Tableau dashboard on the Superstore dataset — comprehensive overview of profit trends across time frames, product categories, and geographies. Designed for executive-level decision making.
Deep dive into 8,800+ Netflix catalog data points identifying viewership trends. Revealed that adding TV Shows in July & December could increase viewership by up to 15% — providing data-driven insights for content allocation strategy.
Identified key customer characteristics influencing treadmill model purchases using outlier detection, correlation heatmaps, and probability analysis. Actionable insights estimated to drive a 10–15% uplift in sales conversions via data-driven product recommendations.
Premium scroll-driven landing page for Porsche 911 Turbo S. Canvas frame-by-frame animation (192 WebP frames via FFmpeg), circle-wipe hero reveal, horizontal marquees with GSAP + Lenis, staggered section animations, dark overlay with counter animations. Zero frameworks.
Browser-based implementation of the classic Simon memory game — sequences of colour and sound patterns that increase in length with each round. Vanilla JavaScript, DOM manipulation, and event-driven game logic with progressive difficulty.
Restaurant web app built while learning React — component-based menu display, interactive UI, and responsive layout. A hands-on project for applying React fundamentals including state management, props, and component composition.
RESTful API built with Node.js, Express, and MySQL for managing candidate recruitment data. Full CRUD operations for candidate profiles, score assignment across three evaluation rounds, and analytics endpoints returning top performers and per-round averages.