Local SLM × Edge Inference for Construction Sites — Our Internal R&D

AIM Inc. has launched internal R&D on an edge inference architecture centered on local SLMs and RAG to meet the construction site’s requirements for low latency and high availability. Aiming for a future where each edge node performs autonomous reasoning in sites operated by unmanned robots, we are advancing prototype implementation and the establishment of evaluation metrics.

Intended Use Cases

• Instant responses for on-site Q&A and work instructions
• Proposals for automatic schedule replanning based on progress signals
• Early detection of safety/quality deviations and corresponding guidance

Technical Approach (Principles)

• Deploy distilled + quantized small language models at the edge
• Use RAG to reference BIM/CAD, schedules, and manuals as external memory
• Ensure fault tolerance and resynchronization to continue operation even during connectivity loss

Progress

• Phase 0: Requirements definition & KPI design (latency, instruction-compliance rate, interruption rate, etc.)
• Phase 1: Prototype implementation (in-situ latency validation underway)

Note The initiatives described on this page are our own research and development and do not imply any contract or joint execution with any specific company.

Contact Book a meeting here: https://calendar.app.google/TZh75w1LU7JJ51V1A