At Gofrret, I built and shipped a drone-based solar panel inspection platform, developing its TypeScript and React workflows for video uploads, job tracking, and panel-level results. I also built resumable S3 uploads and orchestrated GPU jobs with Terraform-managed AWS Batch and Step Functions.
I'm building an LLM inspection assistant that answers natural-language questions about panel-level results through Claude tool calling and read-only, tenant-scoped PostGIS queries. I evaluate it against a benchmark question set, and implemented PostgreSQL row-level security, idempotent job submission, and a transactional outbox.
At FraunhoferIEE, I automated scalable Python training workflows on HPC Slurm clusters for my thesis on correcting short-term wind measurements. My ensemble prediction workflow reduced data collection requirements by four times; earlier, at Inveon, I developed a C# and SQL reconciliation service that reduced order-data discrepancies by 90%.
