Stillwind Logo
Stillwind Logo

Autonomous Electrical Engineering

Our first product, Stillwind Search, makes finding an electronic part as easy as writing a text message. The search engine turns natural language queries into fine-grained specifications that are matched against our proprietary database of millions of parts.

This is the first step towards autonomous electrical engineering.

Our Path to Autonomous Electrical Engineering

Data Foundation

  • Semantic+Exact Component Search

    Current databases of electronic parts like distributor catalogues and datasheets aggregators know very little about their parts apart from the datasheet URL. They use coarse-grained schemas. Additionally, it is not possible to simultaneously search in the databases both semantically and exactly (i.e., a query like "a MCU I can put into a toy rocket with 5V operating voltage" is difficult to exactly match with only embedding searches, etc.).
    Stillwind Search, our search engine currently in development, is designed to address these challenges.

Verification

  • Real-Time Digital Simulation

    For software developers, it's natural to have a tight and fast trial-and-error loop where we can make some changes to our software and test the results instantly. We want the same thing for electrical engineers and embedded software developers.
  • Analog Circuit Modeling

    Real complexity is not only in digital components but also in analog components that have complex dynamics. Research has shown that they can be modeled with modern machine learning methods. We employ these methods to build (reasonably) accurate real-time simulations of analog circuits.
  • Firmware-in-the-Loop

Intelligence

  • Spatial Reasoning

    Language-based foundation models are inherently bad at spatial reasoning, which is necessary for the tasks of placing and routing components. Any reasonable AI hardware engineering system needs a spatial intelligence layer, probably achievable with finetuned energy-based models.