Sonatus

We use AI as a coding assistant on most of our projects. It's not always about adopting AI for a specific task; often, it's about leveraging LLM-based tools to help write software more efficiently. In one recent example, I needed to convert a binary file into a human-readable table.

Reading binary data manually is time-consuming, but using Gemini, I was able to generate a Python script that converted it into a CSV in seconds.

Without AI, that task would have taken much longer and I would have had to manually figure out how to parse and process the binary data. While AI speeds up tasks, it isn't perfect. LLMs can overcomplicate simple problems or generate buggy code that tries to patch symptoms rather than fixing the root cause.

Used correctly, AI dramatically improves productivity, but it's important to remain critical and not become overly dependent on it.

Automotive AI is actually the culmination of software-defined vehicles. Sonatus's mission since its founding has been to accelerate the adoption of software-defined vehicles in the auto industry, enabling vehicles to be flexible, updatable, and dynamic.

With AI, it now allows us to add intelligence to that list and enable compelling new capabilities that were not possible before.

For some context, one of our products, Collector AI, helps OEMs collect vehicle data that enables specific inquiries and evolves them over time through event-driven collection policies. Vehicle engineers can easily create policies through a graphical user interface or, if they want, by directly editing JavaScript Object Notation code.

But these approaches require detailed knowledge of the vehicle by an expert engineer. One monumental moment was when we were also developing an AI enhancement to make it easier for more users inside the OEM, not just technical experts, to use the product.

This enhancement enabled them to describe in natural language what they wanted to collect based on specific trigger conditions. While experimenting, we discovered that the tool could intelligently identify useful, relevant information that exceeded our original expectations when we wrote our queries.

The AI was able to really understand the intent of the query, and we realized that AI could serve as a thought partner to our users, not only helping them to be more efficient but also helping them to do their work better. We are now rolling out this capability as a fundamental new feature of our product.

We collectively read a lot. Everything from academic papers to Medium articles and blog posts to market and industry reports. We have weekly technology team sessions where team members "show and tell" on a rotating cadence over cutting-edge topics, both AI and non-AI.

Team members are encouraged to attend conferences and symposia and the company supports it. The only condition is to provide a trip report presentation to share your learnings with others!