Raj Nakarja spent a decade designing wireless and embedded systems across wearables, industrial IoT, smart home, and consumer electronics. It’s the kind of background that gives you a clear view of what works in connected product development, and more importantly, what doesn’t. The recurring frustration, as Raj describes it, is that every new IoT project ends up rebuilding the same infrastructure: stitching together connectivity, databases, device management, and dashboards, over and over again. The sensors exist. The connectivity exists. But the integration overhead means most projects stall long before they deliver value.

Silicon Witchery is his attempt to collapse that stack. The company’s S2 module is a cellular IoT device that ships with global LTE-M connectivity and a SIM included. No hunting for carriers or negotiating data plans. Devices can be reprogrammed entirely remotely via browser, which means algorithms can evolve in the field without retrieving hardware. For fleet management, data storage, and application logic, everything runs through Superstack, their integrated cloud platform.

The more interesting piece is the AI layer. Rather than building dashboards that require users to interpret data themselves, Silicon Witchery enables natural language queries to real hardware over WhatsApp or REST APIs. Ask a question in plain English (“Are my crops okay?” or “Is the warehouse secure?”) and get an answer back. But unlike typical LLM integrations that send data to external providers and return a best guess, their system generates deterministic algorithms from the query, then runs your data through them locally. The answer is calculated, not inferred, and your data never leaves the server. It’s a meaningful architectural choice for anyone dealing with sensitive operational data.

Hardware, connectivity, cloud, database, and AI, all included, all integrated. For researchers, integrators, or businesses tired of rebuilding IoT plumbing, it’s worth a look.