Senior Machine Learning Engineer

AI search solution across 15+ brand help centers, serving grounded answers and search results from 400+ brand websites and knowledge articles spanning multiple countries, markets, domains, and languages, enabling consumers and healthcare professionals to resolve queries independently, reducing cost and reliance on human agents. The experience feels familiar — an AI answer sits above the search results like a Google AI Overview, and it types out live, word by word, like ChatGPT.

• Built an automated pipeline that keeps content fresh — crawling brand websites, refreshing what changed, and removing outdated pages before each update.
• Processed content from web pages and product leaflets, cleaned it up, and made it searchable through a combination of keyword and meaning-based search.
• Delivered fast search with autocomplete and suggestions, plus an AI assistant that answers questions using only trusted brand content and streams the response in real time.
• Added automated quality checks that continuously score answers for accuracy, safety, and brand compliance across both test cases and real traffic.

AI assistant that lets procurement and business teams ask questions in plain English across 11,000+ Contract IDs and 30,000+ contracts covering ~$7B third-party spend — replacing hours of manual contract review.

• Chat-based assistant that answers questions about contracts — from quick counts and lists ("how many contracts expire in 2026") to specific clause wording like payment, termination, or auto-renewal terms.
• A true agentic system — for each question the assistant reasons about what's being asked, decides which sources to check, digs deeper when the first pass looks thin, and combines the evidence into one grounded answer with follow-up question suggestions.
• Two ways to work: a fast chat mode for everyday questions, and a Deep Research mode that plans the angles worth investigating, sends out specialised agents to study many contracts in parallel, and compiles everything into a structured, downloadable report.
• Answers stream in live with visible progress ("understanding your query", "searching contract documents"), and users can narrow the scope with filters for category, supplier, or a specific contract.
• Delivered faster contract review, clearer compliance visibility, and measurable time and cost savings across procurement and business teams.

Commercial Tech — Marketing Mix Models

MLOps framework for Marketing Mix Models using Databricks Bundles, Unity Catalog, and GitHub Actions for scalable, repeatable modeling of brand and market-level sales drivers. Built robust CI/CD workflows with linting, bundle validation, pre and post semantic versioning, and automated Databricks job orchestration across Dev/UAT/Prod. Enabled dynamic, market and brand specific task generation at runtime, versioned model and data persistence in Unity Catalog, reproducible deployments via explicit semantic release tags, with outputs seamlessly refreshed in Power BI for stakeholders to optimize global marketing spend.

MLOps and Agentic-RAG Foundation

Comprehensive MLOps template for Databricks leveraging GitHub Actions to automate CI/CD pipelines, including code linting, testing, environment setup, and seamless deployment of Spark jobs, such as model training, validation, and deployment, resulting in significantly enhanced workflow efficiency and reduced deployment times. Delivered standardized framework code and detailed documentation, reused across Help Center, Contract AI, NRM, NBA, QSC, and Commercial Tech projects.

Haleon GenAI Assistant

Architected and developed an enterprise-grade, multi-team GenAI platform integrating FastAPI, React, Azure OpenAI and Azure Cloud Services to enable secure, intelligent data access across Finance, Procurement, and Help Center from scratch.

• Built team-specific AI agents and modular tool frameworks using LangChain and LangGraph for contextual data retrieval.
• Enhanced retrieval quality through semantic ranker with multilingual support and hybrid search using text and vector search in Azure AI Search.
• Engineered a scalable, structure-aware document processing pipeline using Azure AI Document Intelligence with automated GitHub Actions workflows.
• Experimented with MCP (Model Context Protocol) client–server architecture for distributed tool orchestration across databases and AI services.
• Optimized chat architecture for sub-second latency through asynchronous processing, containerized deployment, and scalable microservice design.

Asset Vision

Designed and implemented a robust CI/CD pipeline with GitHub Actions to build and publish Docker images to GitHub Container Registry (GHCR) and Azure Container Registry (ACR), enabling automated deployment to Azure Web App and microservices on Kubernetes cluster for scalable, secure and reliable application management across environments. Applied same approach to streamline deployments in other RAG based projects.