Featured Projects

Sparky AI — Audience Intelligence (M&S)

Building an AI-powered audience selection tool at M&S using Databricks AI Agents and LLMs. Sparky helps business users identify the right target audience for Sparks offers and promotions — translating plain English queries into governed data queries against loyalty programme data. Also building an interactive AI chatbot for internal BU users. Demonstrates real-world AI agent architecture: natural language → structured query → governed output.

Mission Desk — M&S Sparks Loyalty Platform

Solo-built full-stack enterprise platform managing 1M+ Sparks loyalty mission records at Marks & Spencer. Real-time validation against 15+ business rules, 6 major releases, 10K+ lines of code. Built with AI-assisted development (Copilot/LLMs). Reduced deployment time by 97% (7 days → 4 hours), cut data quality errors by 85%, delivered £500K+ annual business value. Nominated at company-wide Sparks All Hands. Part of the team that shipped the new personalised Sparks programme (April 2025).

SevOne Code Generator Framework

Built a Python/Jinja2-based code generation framework at Sky UK that automated data pipeline creation for network telemetry ingestion from SevOne IBM (on-prem) into BigQuery (GCP). Reduced new data acquisition onboarding time by 95% and manual code conversion by 90%. Enabled teams to provision pipelines in minutes instead of days with zero configuration drift.

Real-Time Kafka Analytics Platform

Architected a high-throughput Kafka streaming pipeline at Sky UK processing millions of network events daily using Apache Kafka, kSQL, and Kafka Connect for real-time ingestion from SevOne IBM to BigQuery. Built a Kafka Connect Task Monitoring Tool and BigQuery Monitor CLI. Impact: 30% reduction in operational latency, 95% reduction in pipeline downtime.

NLP-Powered Email Lead Bot

Built an AI-powered email analytics service at Gartner using NLP that automatically identified and qualified sales leads from 10K+ emails monthly. Saved the sales team hundreds of hours while identifying 200+ qualified leads per month. Resulted in a $700M revenue impact by streamlining renewal processes and improving renewal rates by 35%.

AWS Serverless Clinical Platform

Led the cloud modernization of Eli Lilly's Clinical Supply Platform, migrating legacy systems to AWS serverless architecture. Led a team of 5 engineers, implemented event-driven pipelines processing 100,000+ messages/day with Lambda, Step Functions, IBM MQ, and S3. Impact: 30% infrastructure cost reduction, 25% operational performance improvement.

Production-minded Retrieval-Augmented Generation system built governance-first: PII masking on every query, structured audit logging, heuristic confidence scoring, and source attribution. Stack: Claude (Anthropic), LangChain, ChromaDB, sentence-transformers, Streamlit. Demonstrates that enterprise AI adoption depends on trust and traceability, not just model quality.

This portfolio website built with Next.js, Tailwind CSS, and MDX. Features dark mode, technical blog posts, and responsive design. Demonstrates modern frontend development practices and serves as a platform for sharing AI engineering and data platform insights.