Aron Metzig

I extended open source libraries with batching support and optimized it for NVIDIA H100 clusters using vLLM and custom CUDA kernels, achieving 150% higher throughput than the baseline.

I unified distributed state using Redis as the central source-of-truth, and decoupled the real-time transcription pipeline via async messaging. Supported by full-stack Grafana observability, this fault-tolerant distributed messaging eliminated REST bottlenecks and ensured zero data loss during high-load WebSocket streaming.

Leveraging my background in data science and low-level systems, I overhauled the applied AI pipeline. By integrating sub-symbolic models and clustering algorithms for live speaker diarization, and extending open-source components with Rust and custom CUDA optimizations for NVIDIA H100 clusters, I achieved 150% higher throughput than the baseline, due to a custom batching strategy.

For strict regulatory compliance, I engineered GDPR-aligned recording consent flows alongside cryptographic client-side watermarking and tamper protection.

Architecting a BaFin-regulated FinTech platform from scratch, I engineered a highly available data orchestration layer. I designed a zero-downtime central scheduler to ingest corporate actions, FX rates, and equity quotes across multiple brokers, utilizing Raft consensus for distributed HA state management.

To decouple brittle broker integrations, I implemented an asynchronous, AMQP-backed messaging backbone guaranteeing "at least once" transactional persistence via PostgreSQL. I fortified this distributed messaging architecture with a full-stack Grafana, Prometheus, and Loki observability pipeline to ensure proactive incident detection and strict uptime SLAs.

I codified the estate in Terraform with VPN-isolated clusters, cutting infrastructure costs by 20%. I also authored reusable Keycloak authentication libraries, standardizing RBAC and audit logging across all services. Greenfield indices service went from first commit to production in under one week.

Operating as the lead architect and requirements engineer for a B2B/B2C timber business, I partnered directly with stakeholders to translate complex logistics and pricing rules into a clear technical roadmap. I then designed their enterprise platform from the ground up, bridging the gap between business needs and system architecture.

I built a centralized PIM catalog for 10,000+ SKUs featuring automated vendor syncing, AI-driven clustering, and a custom Studio UI tailored to the internal content team's workflows. To streamline back-office operations, I engineered an asynchronous ERP sales-order integration alongside a dynamic logistics rules engine orchestrating DPD, UPS, and internal freight fleets.

Focusing strictly on performance and conversion, I deployed a custom Nginx edge-caching layer and an optimized image pipeline, slashing page load times by up to 235%. Combined with automated technical SEO infrastructure, this robust architectural foundation directly drove a 280% increase in annual revenue.