RF Sensing Is Quietly Becoming a Product Category: What QuadRF Signals for Builders
A hobbyist device that spots drones and detects WiFi signals through walls shows how far radio-frequency sensing has come. For business owners, the interesting part isn't the hardware. It's what happens when you pair cheap sensors with AI that can actually interpret the signal.
A gadget that sees through walls
Jeff Geerling recently put a device called QuadRF through its paces, and the results are the kind of thing that used to require a government budget. It can detect drones flying nearby and pick up WiFi activity through the walls of his house. The radio spectrum is full of information we can't see, and tools like this are making it readable by regular people.
That's a fun demo. But if you build products for a living, the demo isn't the story. The story is the pattern underneath it.
Cheap sensors are outpacing our ability to use them
Hardware like QuadRF is part of a broader shift. Sensors that capture the physical world (radio frequency, audio, vibration, temperature, motion) keep getting cheaper and more capable. A single device can now stream more raw data than a person could ever review by hand.
And that's the gap. Collecting signal is close to solved. Interpreting it is where the value sits, and where most projects stall.
A raw RF capture is just noise until something decides what matters:
- Is that a drone or a passing car?
- Is this WiFi pattern normal, or does it suggest a device that shouldn't be on the network?
- Which of these ten thousand readings deserves a human's attention right now?
Those are classification and pattern-recognition problems. They're exactly the kind of work modern AI models are good at, provided someone builds the pipeline that connects the sensor to the model to a decision.
Where AI turns signal into a product
I've spent a lot of my career on the unglamorous middle layer: taking a stream of messy real-world data and turning it into something a business can act on. When you apply that lens to a device like QuadRF, a product roadmap starts to appear.
Consider the categories of software this hardware invites:
- Detection and alerting. An AI agent watches the RF feed and only pings a human when something crosses a threshold that matters. No dashboards to babysit.
- Baselining and anomaly detection. The system learns what "normal" looks like for a given space over a few days, then flags deviations. This is far more useful than static rules, because every environment is different.
- Enrichment and context. Pair the sensor data with other sources (schedules, camera feeds, access logs) so the AI can reason about why a signal appeared, not just that it did.
- Natural-language reporting. Instead of forcing a security manager to read spectrograms, an agent summarizes the day: "Two unrecognized drones near the loading dock, both between 2 and 3 PM."
The hardware answers "what is happening in the air right now." The software you wrap around it answers "and what should we do about it." That second question is the one people pay for.
The real lesson for entrepreneurs
Every few months a new sensor or open-source tool lowers the cost of capturing some slice of the physical world. Most of them get written up, admired, and forgotten. The ones that turn into businesses are the ones where somebody built the interpretation layer on top.
That layer is rarely one big model. In practice it's a set of smaller pieces working together:
- A reliable ingestion path that doesn't drop data.
- A model (or a few) tuned to the specific signals you care about.
- An agent that decides what's worth escalating.
- An interface a non-technical person can trust.
Getting those pieces to cooperate is the actual engineering, and it's where a lot of AI projects go sideways. It's tempting to bolt a language model onto a data feed and call it done. Production systems need guardrails, monitoring, and a clear sense of what happens when the model is wrong.
If you're sitting on a stream of data
Maybe it isn't RF. Maybe it's transaction logs, support tickets, IoT readings from your equipment, or a firehose of customer messages. The shape of the problem is the same: you're collecting more than you can use, and the value is locked up in interpretation.
That's the work I do. I build the AI agents and automation systems that sit between raw data and a decision, so the signal you're already capturing turns into something your team can act on.
If you've got a data source that feels underused, or a manual review process that's eating your team's hours, that's usually a good place to start. I'm happy to talk through what's feasible and what it would take to build.
