In this fireside chat, Mircea Neagovici (Head of AI and ML at UiPath) walks through how UiPath actually builds, deploys, and improves enterprise-grade AI.
This is not marketing — it’s a candid, technical discussion about real-world constraints, tradeoffs, and research directions.
What This Video Covers:
- How UiPath builds reliable AI Agents and Document Understanding in production
- Frontier vs open-source models and when each is the right choice
- Why continuous learning + human-in-the-loop is critical after deployment
- Fine-tuning at scale: global models vs per-customer / per-agent models
- Why UiPath uses LoRA adapters for efficiency and enterprise scalability
- Supporting cloud, VPC, and on-prem AI for regulated environments
- How agents improve over time: prompt learning → in-context learning → fine-tuning
- Why reinforcement learning is the next step for DU and Agents
- Computer Use: what’s production-ready today vs what’s still hard
- Modern DU beyond OCR using vision + layout + LLMs
- Benchmarking real enterprise apps with UI Cube
- What’s next: coded agents and machine-first APIs
If you care about reliable AI Agents, DU, and real-world ML tradeoffs (not hype) —
this video is required viewing. !