12 projects
Internal AI Data Analyst built with Gemini, Haystack, FastAPI and MCP servers. Allows team members to query Metabase dashboards, PostHog analytics, and BigQuery via natural language.
End-to-end production chatbot using Retrieval-Augmented Generation for a major Malaysian media group. Deployed on GCP with FastAPI and Haystack, serving customer queries across multiple digital channels.
Personalised learning plan generator for an edtech platform. Uses a Bayesian-based recommendation model to adaptively sequence content based on student performance and learning history.
Hyper-personalisation recommendation engine that combines user geo-location signals with behavioural data to drive organic article engagement across media properties.
Refactored a production data warehouse to follow medallion (bronze/silver/gold) architecture using BigQuery, dbt and Fivetran. Improved pipeline reliability, test coverage, and documentation.
Led migration of legacy on-premise ML inference services to GCP with containerised deployments, Cloud Run and monitoring via Google's Vertex AI.
Designed and monitored A/B testing experiments to evaluate NLU model variants, reducing inference cost and improving intent classification accuracy.
Migrated Airflow-orchestrated ETL pipelines and dbt models from AWS Redshift to Snowflake, ensuring zero-downtime cutover and full data validation.
NLP project mining millions of Steam game reviews, performing data validation and preprocessing, then applying neural networks and traditional ML for aspect-based sentiment analysis.
Delivered multiple client-facing POCs spanning computer vision, NLP and forecasting for DataMicron Systems. Scoped, built and handed off end-to-end solutions including post-launch support.
Maintained and extended a Django-based internal web application used for hardware QA testing workflows on Android TV firmware releases.
Personal portfolio site built with FastAPI, Jinja2 templates and Tailwind CSS. Zero JavaScript frameworks, GitHub Dark design, containerised with Docker.