I Make Businesses Findable in the AI Era
I build websites that AI engines read and cite, plus the automations that take over what happens after someone contacts you.
I started in classic web development. Since then, how people search has changed: more and more of them ask an AI directly and get a single answer with two or three names. If yours isn't there, you don't exist for that customer — no matter how well you rank on Google. That's what I fix.
"You work directly with me — no middlemen, no outsourcing. Clear communication, real delivery, real results."
Darius Osadici
AI Visibility & Web Development
How I Work
Results-Driven
Every line of code I write is focused on delivering measurable business outcomes. I don't just build features—I solve problems.
Optimised For How People Search Now
Not just for Google. For ChatGPT, Claude, Gemini, Perplexity and AI Overviews — where your customers already are.
Clean Code
I write maintainable, scalable code that your team can understand and extend easily. No shortcuts, no technical debt.
AI-First Thinking
I always look for opportunities to leverage AI for automation and efficiency gains, giving you a competitive advantage.
Technical Expertise
Visibility & Structured Data
- Schema.org / JSON-LD
- Answer Engine Optimization
- Generative Engine Optimization
- llms.txt
- Google AI Overviews
- SEO tehnic
AI & Automation
- OpenAI / GPT-4
- LangChain
- n8n / Make / Zapier
- Custom AI Agents
- Chatbots
- OCR & Document Processing
Frontend
- React / Next.js
- TypeScript
- Tailwind CSS
- Framer Motion
- Responsive Design
- UI/UX Best Practices
Backend
- Node.js
- Python
- PostgreSQL / MongoDB
- REST & GraphQL APIs
- Supabase / Firebase
- AWS / Vercel
Tools & Workflow
- Git / GitHub
- CI/CD Pipelines
- Agile Methodology
- Technical Documentation
- Performance Optimization
- Security Best Practices
Education & Research
BSc in Computer Science and Information Technology
Faculty of Automation and Computers, Politehnica University of Timișoara
AnchorDepth
Self-supervised monocular depth estimation
I adapted Depth Pro — Apple's 952-million-parameter model — using LoRA and a consistency anchor loss.
The problem is a familiar one: fine-tune a strong zero-shot model on unlabeled video and you usually end up worse than you started. AnchorDepth anchors the adapted predictions back to the original zero-shot outputs, so the model keeps what it already knew while still learning from the new data.
It improves on Depth Pro across KITTI, Cityscapes and Make3D, without a single ground truth label. Trained end to end on one RTX 4070 Ti with 12GB of VRAM.
Supervisor: Prof. Dr. Habil. Ing. Călin-Adrian Popa
Why this matters for your project
I don't sell "AI" as a buzzword. I've worked inside large models, under real resource constraints, on a problem where most attempts fail. That means I know what these models can do — and, more importantly, what they can't. Including when someone else promises you something that doesn't work.
Ready to Start a Project?
Whether you want to show up in AI answers, get rid of manual work, or both — let's see where you stand.
Get in Touch