Turning chip datasheets into AI-structured workflows for embedded engineers

Integrating hardware and software remains one of the biggest challenges in embedded development — from navigating complex datasheets to adapting designs under shifting supply chain constraints. London-based startup Embedd, co-founded by hardware entrepreneur Michael Lazarenko, is addressing this with an AI-driven platform that converts chip documentation into structured, machine-readable models. The result: vendor-agnostic tools that allow engineers to configure devices, generate deterministic code, and integrate seamlessly into existing workflows. eeNews Europe talks to Michael Lazarenko about the origins of Embedd, its mission to apply AI for hardware–software integration, and the opportunities it opens across industries from consumer to automotive and aerospace.

Michael Lazarenko, CEO and co-founder of Embedd

eeNews Europe: What inspired you to co-found Embedd? What problem were you most determined to solve in the embedded systems industry?

We started Embedd after a decade of facing hardware–software integration challenges ourselves. Before Embedd, we built a wireless sensing company and took multiple products from idea to mass manufacturing. We saw firsthand how difficult it is to integrate chips, stay innovative, and add new capabilities — and how helpless you can be when supply chain issues hit, forcing you to rework an entire firmware project just to achieve support. The war in Ukraine took our production facility, and that was the last straw. We decided to turn the table and build for builders — changing how hardware and software come together.

Embedd’s vendor-agnostic solution for MCU configuration and device-tree generation accelerates setup with broad MCU support or custom DTS uploads.

eeNews Europe: Your mission is to make chip datasheets machine-readable with AI. How does this change the daily workflow for embedded engineers?

Our thesis is that hardware–software integration is mostly a data challenge, not a software development challenge. We use AI to retrieve and structure data, then give engineers convenient ways to access and control it through a simple UI. Every datapoint is back-linked to the datasheet, so engineers always see the full context. Importantly, they stay in control — configuring the data and generating deterministic, clean, and documented code. This lets engineers focus on higher-level application challenges while we make the hardware abstractions agile and flexible. And because our tooling is vendor-agnostic, you can configure MCUs from different manufacturers within the convenience of a single platform — something that’s never been possible before.

Embedd’s MCU configurator transforms raw specs into an intuitive interface for configuring devices, generating drivers, and reducing setup errors.

eeNews Europe: Where do you see the biggest opportunities for AI in embedded development today — and what challenges still remain?

It depends on the software layer and the industry. At the lower level, our core thesis is that AI should be used to resolve data challenges, not to generate code directly. At the application layer, AI can potentially accelerate a lot of mundane development tasks. But industry context matters: in functional safety domains, AI-generated application code is still only useful in rare cases. For other applications, it has a potential of becoming an accelerator.

Embedd’s deterministic generation engine delivers production-ready code and documentation with functional safety and enterprise-grade reliability.

eeNews Europe: Which sectors stand to benefit most from Embedd’s approach, and why?

Industries with more constraints stand to gain the most. A small consumer startup has flexibility to choose parts and architectures while not worrying too much about compliance. But in automotive, defense, medical, or aerospace, you often have fixed architectures and part choices, limited existing support, bigger compliance burdens, and longer cycles. That said, we see demand across the board — from consumer to automotive to defense.

eeNews Europe: What feedback have you received from early users or partners, and how has that influenced your roadmap?

Our early focus was on supporting peripheral devices. But the overwhelming feedback we got was about MCUs — the same challenges, but at a much higher level of complexity. Users also pushed us to go beyond bare-metal to support Zephyr, Linux, and AUTOSAR.

eeNews Europe: Your system parses datasheets, errata, and SVD files into structured data models. What approaches do you use to handle inconsistencies, ambiguities, or missing details in vendor documentation?

For each device class, we build an internal intermediate representation — it defines the required data points and maps class-specific functions to them. This gives us a clear view of the completeness of the documentation. We then run automated checks, and in most ambiguous situations, have human-in-the-loop checks. Wherever possible, we prioritize deterministically extracted and user-validated data, so the models stay robust and reliable.