I lead the technical direction and hands-on development of Hisani, an early-stage horse-racing platform. I am building the application from the ground up, with work spanning the stopwatch, race planner, marketplace, and stable-management modules. My responsibilities cover application architecture, feature implementation, authentication, administration, and delivery workflows. I translate product requirements into connected functionality and work across the platform’s technical implementation. Hisani is currently in development.

I built RacingStatto end to end and lead its technical development, covering the data-processing pipelines, ranking and combination-score systems, and user-facing application. The platform brings historical racing data into daily racecards, runner comparisons, and searchable form analysis. I also developed filtering, daily data updates, and membership functionality, including trial access and paid subscription tiers. My role spans implementation and technical decision-making across the product. RacingStatto is now live with early test users.

I research online learning for deep models in changing data streams, focusing on how historical examples can support training under memory and latency constraints. My work centres on an Event-Conscious Historical Oversampling operator that selectively reintroduces historical examples during training. Using Python and River, I am developing hardness-weighted replay policies and scheduling mechanisms for delayed labels, concept drift, and changing stream throughput. The research examines how to allocate limited training resources as the data and availability of feedback evolve.

I developed VitAI and GlitchWitcher during Google Summer of Code 2025, working on AI-assisted repository analysis and software defect prediction for Eclipse Adoptium. VitAI combines an MCP interface, repository exploration, and retrieval-augmented generation to support context-aware questions and test triage. For GlitchWitcher, I implemented the published REPD method and supporting data-processing and analysis workflows. My contributions included repository indexing, model-related pipelines, and GitHub Actions integration, with work merged into the upstream Adoptium repositories.

I led a regional survey on Veritas Backup Exec across the Middle East and Africa and analysed the responses to produce sales-relevant insights. The assignment involved interpreting the collected data and translating the findings into information that could support sales activity. This gave me early experience applying analysis to a commercial question and communicating findings beyond the technical work itself.