QLoRA fine-tuning experiment on GitHub (opens in a new tab)
DucensAI started as a marketing agency and became an AI automation company. I was the CTO from January 2024 to April 2026.
AI invoicing for a B2B hardware client. The system reads the client's records, makes the invoices, finds service periods that are about to expire, and sends every exception to a person as a task with a link to fix it. It saves the team 10+ hours a week.
Private legal research assistant. A local RAG system (retrieval-augmented generation: the model answers from documents it searches first) over a mid-sized law firm's internal case documents. It returns exact quotes, plain-language summaries and the case names in seconds.
Security automation. A proof of concept for automated security work.
Fine-tuning experiments. A QLoRA fine-tuning run on a local RTX 3080 GPU. The script and dataset are on GitHub.
Before the AI work, DucensAI (then Ducens Media) ran email marketing for small businesses. My best result -> a holiday email campaign for a car detailer that added $2,520 in his off-season (the original page called it 504% seasonal revenue growth).