Notes from building Physical AI.
What we learn building cognitive systems, training small and multimodal models, collecting real-world data and deploying AI on devices.
Fine-Tuning Is Dead. Long Live Fine-Tuning.
Prompt engineering took us far. But for domain-specific production workloads, parameter-efficient fine-tuning is back.
Apr 17, 2026 · 12 min read
Why Latency Is a Product Feature, Not an Engineering Detail
Users do not tolerate spinners. In finance and healthcare, 400ms is the difference between adoption and abandonment.
Apr 17, 2026 · 6 min read
RAG Beyond Retrieval: Structured Context for Enterprise LLMs
Plain vector search falls apart at scale. How XoidLabs combines graph retrieval, semantic filters, and policy-aware context.
Apr 17, 2026 · 10 min read
Building Production-Grade AI Agents at XoidLabs
How we architect deterministic AI agent systems for enterprise workloads, balancing autonomy with auditability.
Apr 17, 2026 · 8 min read
Your Next Machine Shouldn't Just Operate. It Should Understand.
Let's Engineer What Comes Next.
You bring the product idea. We bring the intelligence, data and engineering to make it work.