Natalia – Python, OpenAI, LangChain

An AI-powered immersive 3D learning platform for children aged 7–14, available to families as a freemium app and to schools as a B2B product. Kids explore interactive 3D worlds, solve curriculum-aligned challenges, and learn through play across a range of subjects. Core features include an AI teaching assistant that guides learners through in-world interactions and curriculum-aligned content generated from structured knowledge sources. Parents and teachers get real-time progress tracking and visibility into learning outcomes, making it a practical tool for both home and classroom use.

• Architected the end-to-end AI system from the ground up, including lesson and course generation pipelines, adaptive learning engine, RAG retrieval, teaching assistant, and hallucination reduction framework;
• Designed and owned the full backend and API architecture, including auth systems, database design, and cost-aware LLM querying;
• Built and maintained multi-protocol API infrastructure (REST, Socket.IO) supporting real-time AI interactions across the platform;
• Implemented production-grade reliability practices, including feature flags, monitoring, and deployment pipelines;
• Led and delegated across frontend, backend, and immersive engineering teams, owning technical alignment and unblocking delivery across all streams;
• Ran structured client delivery cadence, including weekly demos and async stakeholder updates, translating product requirements into executable technical solutions.

A B2B logistics and fleet tracking platform serving major carriers, including Royal Mail and DPD. The platform aggregates and processes large volumes of operational data across distributed fleet networks to deliver real-time visibility, performance analytics, and business intelligence to enterprise clients. This role focused on building the internal data infrastructure needed to consolidate fragmented company data into a centralised, observable platform by laying the technical foundation for scalable MLOps and ML lifecycle management.

• Designed and implemented a centralisation schema to unify 5 disparate data sources into a single ELK stack, enabling scalable real-time ingestion and observability across millions of records;
• Engineered data pipelines capable of handling high-volume operational fleet data with reliability and consistency;
• Partnered with DevOps to implement monitoring and logging infrastructure across the platform;
• Prepared the architecture for full ML lifecycle management, establishing the technical foundation for future MLOps workflows.

An edge AI company delivering real-time video intelligence solutions for physical security, smart cities, and retail analytics. The core product runs on-device AI inference on existing camera infrastructure, enabling features such as people detection, tripwire violation alerts, occupancy counting, and privacy-preserving face blurring for shops and businesses. The platform was built to run efficiently on edge hardware, minimising latency and cloud dependency for real-time deployment environments.

• Architected GPU-optimised AI inference pipelines leveraging NVIDIA CUDA for real-time video processing across people detection, tripwire detection, occupancy counting, and face blurring features;
• Developed an anomaly detection model achieving 97% accuracy for abnormal device behaviour, including camera dismount detection;
• Built a pre-ChatGPT video-to-text matching system using OpenAI CLIP, enabling natural language queries against video footage with exact timestamp retrieval;
• Designed and launched a flagship product on the NVIDIA Omniverse Extension Library, passing quality control without major revision and achieving several hundred public downloads;
• Led cross-functional delivery across 3D, graphics, backend, UI, and marketing teams from development through public launch;
• Partnered directly with NVIDIA, attracting a ~$20M telecom investment interest from Saudi Arabia off the back of the Omniverse product.

A global semiconductor manufacturing company. The role was part of a scholarship programme funding the M.Sc. in AI, embedded within the company's manufacturing operations to design and deliver production-grade AI applications that address real operational challenges on the factory floor. The platform grew from an initial predictive maintenance system into a broader suite of AI tools spanning multiple areas of the semiconductor manufacturing pipeline.

• Designed and delivered a predictive maintenance platform generating at least €250K in revenue from the first production deployment, with expansion underway across additional manufacturing domains;
• Reduced equipment downtime from 45 minutes to 15 minutes across 300+ machines through integrated predictive modelling and real-time operational forecasting;
• Engineered custom Big Data ingestion pipelines processing high-volume industrial logs (~1GB per file, 500+ parameters) under significant infrastructure constraints;
• Built full-stack live dashboards and automated performance reporting with daily and weekly segmented distribution;
• Developed a neuro-symbolic predictive maintenance framework combining deep learning and reasoning models through a knowledge-base middle layer, achieving 85% accuracy on semiconductor manufacturing data (M.Sc. thesis).