Edge AI tiered inference platform based on RISC-V + LiteOS · Full AI in 128KB SRAM · 3-tier progressive inference architecture
智能终端AI中间件
Edge AI tiered inference platform for IoT / industrial embedded scenarios
IoT devices face a “last mile” problem after deployment: device state is opaque, O&M costs are high. Engineers need 30+ minutes per visit to connect debuggers, check logs, and build tools.
Smart Terminal AI fills the gap between “can’t run models on-device” and “pure cloud is unreliable” — letting every MCU device answer “what’s your current status?” in natural language.
The inference scheduler automatically selects the optimal path based on query complexity + network state. Tier 3 failure auto-degrades to Tier 2, ensuring offline availability.
Portable to any RISC-V/ARM MCU + RTOS. Unified ai_engine_query() API with automatic 3-tier scheduling internally.
Intent-aware CLI framework supporting mixed natural language and traditional commands. History + TAB completion. Zero dependency, fully portable.
Standardized CPU / memory / temperature / network / sensor collection interfaces, adapted to mainstream embedded platform drivers.
Lightweight inference acceleration IP for Anlogic EG4X20. Maps matrix multiply to FPGA lookup tables. Theoretical 50x speedup, compressing inference latency to under 200ms.
Plug in USB → Open serial terminal → Ask in natural language:
> ai is the device status normal?
→ CPU usage 12%, memory 21% (128KB/606KB), temperature 36.7°C normal,
WiFi signal good (-58dBm), sensor data collection normal.
Device is healthy.
> ai any anomalies in recent logs?
→ [AI Tier-3] Analyzing recent 47 log entries...
Found 2 WARN level logs: ADC channel 2 signal is 0,
possibly sensor disconnected. All other logs are INFO level,
system operating normally.
Zero App install · Zero special tools · Serial direct connect · Second-level response