Like most of you, I was intrigued by Karpathy's LLM OS architecture. As a result, I created my own version that seeks to solve an interesting problem, making i2c sensors hot-swappable using LLM as an intelligent interface. My purpose to post to this forum is to see whether other people also find this problem and solution as interesting as I. Please give your thoughts and opinions on this. Here is the quick abstract.
Sensor interfacing in embedded systems traditionally requires manual driver development, firmware recompilation, and is limited to static configurations, making dynamic sensor interfacing a complex and inflexible process. This paper introduces OTTER OS, a novel
embedded operating system that leverages Large Language Models (LLMs) and retrievalaugmented generation (RAG) to enable prompt-driven, runtime i2c sensor interface. OTTER OS integrates an intelligent LLM pipeline for datasheet parsing, a domain-specific
intermediate language (OTTER Embedded Language or OEL), and a multithreaded execution engine (OTTER Engine) built on Mbed OS to dynamically instantiate sensor threads from natural language commands. The system supports plug-and-play functionality, multiple concurrent sensors, and reading of i2c data in physical measurements, all without requiring manual code updates or recompilation. Experimental results show that OTTER OS achieves an 80.5% success rate in end-to-end sensor interfacing and 94.9% accuracy in
context validation. OTTER OS’s modular architecture and runtime adaptability demonstrate a new paradigm in embedded sensor interfacing, bridging AI-driven reasoning with low-level hardware control to reduce development time, improve flexibility, and democratize sensor access for non-experts.
Like most of you, I was intrigued by Karpathy's LLM OS architecture. As a result, I created my own version that seeks to solve an interesting problem, making i2c sensors hot-swappable using LLM as an intelligent interface. My purpose to post to this forum is to see whether other people also find this problem and solution as interesting as I. Please give your thoughts and opinions on this. Here is the quick abstract.
Sensor interfacing in embedded systems traditionally requires manual driver development, firmware recompilation, and is limited to static configurations, making dynamic sensor interfacing a complex and inflexible process. This paper introduces OTTER OS, a novel embedded operating system that leverages Large Language Models (LLMs) and retrievalaugmented generation (RAG) to enable prompt-driven, runtime i2c sensor interface. OTTER OS integrates an intelligent LLM pipeline for datasheet parsing, a domain-specific intermediate language (OTTER Embedded Language or OEL), and a multithreaded execution engine (OTTER Engine) built on Mbed OS to dynamically instantiate sensor threads from natural language commands. The system supports plug-and-play functionality, multiple concurrent sensors, and reading of i2c data in physical measurements, all without requiring manual code updates or recompilation. Experimental results show that OTTER OS achieves an 80.5% success rate in end-to-end sensor interfacing and 94.9% accuracy in context validation. OTTER OS’s modular architecture and runtime adaptability demonstrate a new paradigm in embedded sensor interfacing, bridging AI-driven reasoning with low-level hardware control to reduce development time, improve flexibility, and democratize sensor access for non-experts.