Xisiid Intelligence raises $3.7M seed for brain-inspired AI
What's the deal? Singapore-based AI neo-lab Xisiid IntelligenceDealroom has a profile for this one. Try Dealroom → has raised a US$3.44M seed round to fund its Brain-Inspired Self-Evolving Foundation System. The money will also support joint development of AI applications with domain experts.
What's the endgame? The company is building AI that learns from real-world task execution and continuously builds reusable capabilities. Its approach, called Brain-inspired Intelligence for Artificial Intelligence (BI4AI), draws on how biological brains organise cognition, memory, and learning.
What's the product? Xisiid made its public debut at the Shanghai Pujiang Innovation Forum, jointly hosted by China's Ministry of Science and Technology, where it unveiled its foundation system. The system uses a dual-system architecture: a fast execution layer for frequent tasks and a structured memory layer.
Why now? As AI agents move beyond chat into complex professional workflows, evaluation is shifting from benchmark reasoning toward a harder question — can AI reliably complete real-world tasks? In fields like investment banking, consulting, and cross-border legal work, a single assignment can require processing heterogeneous documents and reasoning through many verification steps.
What could go wrong? Current large language models still struggle with long, complex tasks, remaining prone to reasoning drift and compounding errors. Recent Harvey LAB results show leading models complete end-to-end legal tasks less than 20% of the time under its strictest standard.
The founder's view: "Increasing model parameters alone cannot solve the challenges of reliable execution and cumulative learning," said founder Dr. Chua Yam Song. He described the shift as "moving AI from a computation-centric paradigm toward a memory-centric one."
The signal: The round is small, but it reflects a growing bet that the next frontier in AI is not bigger models but reliable execution over long, multi-step work — and that memory, not scale, may be the missing piece.
Read more: prnewswire.com
Image credit: viegas
