Work

JPMorganChase · London, UK

• ◆ Built a production RAG system for Chase UK's mobile banking FAQ assistant: async Python 3.13 monorepo with Litestar / Strawberry GraphQL API, HNSWLib vector search, LLM-driven re-ranking, and Azure OpenAI answer generation over HTTP/2.
• ◆ Engineered multi-layer content-safety guardrails: Presidio PII/PCI detection, LLM intent analysis, and risk evaluation; maintained >80% test coverage with pytest-asyncio, Vitest property tests, and k6 load benchmarks.
• ◆ Delivered the React 19 / Next.js 15 frontend with Apollo GraphQL; managed AWS infrastructure via Terraform (EKS, S3) and Helm.
• ◆ Optimised inference cost by migrating from GPT-4.1 to GPT-5.1 without accuracy regression; first in the ML team to hold a hybrid role spanning traditional ML and full MLOps infrastructure.

Alpine F1 Team · Enstone, UK

• ◆ Led pit-signal accuracy project: applied advanced statistical techniques and error-function minimisation to improve signal accuracy by 91%; results adopted directly by Strategy and Race Track teams.
• ◆ Conducted process mining of 3DX engineering logs, identifying 80% of process divergences across three model years; findings adopted by Enterprise Architecture and PLM teams.

Bosch · Cluj-Napoca, Romania

• ◆ Designed an ANN-based object detection system for automotive front-facing sensors and deployed a Random Forest signal classifier to production.
• ◆ Developed software architecture for an LED Matrix, integrating hardware and application layers.