Standing on a busy range with a student, you don't have 30–60 seconds to wait for a video to upload, process in the cloud, and return a result. By then, the moment's gone, the feel has changed, and the lesson has drifted into small talk.
Edge compute is our answer to that lag. Instead of sending every frame to a distant server and hoping the network cooperates, we run as much of the heavy lifting as possible close to the golfer—on the device itself or on a nearby edge node. That shift changes the entire feel of coaching with technology: feedback stops being "something you wait for" and becomes something that is just there.
What we mean by “edge” in a golf context
"Edge compute" is a broad term, but in golf coaching it really boils down to this:
- On-device processing on the coach's and player's phones or tablets.
- Near-range processing on lightweight edge services close to the user (e.g., regional servers or range networks), rather than a single far-away cloud region.
The goal isn't to eliminate the cloud. The goal is to put the time-sensitive work—pose estimation, swing detection, rep counting—where it can happen fastest and most reliably, and use the cloud where it shines: storage, long-term stats, and coordination across teams and devices.
Why latency breaks coaching flow
Golf coaching is highly temporal. You're reacting to what just happened, while the feel is fresh in the player's body. If technology adds friction at the wrong moments, it stops being a tool and starts being a distraction.
High latency creates a few consistent problems:
- Broken feedback loops: When it takes too long to see a swing with overlays or metrics, players stop trusting the system and default back to verbal-only cues.
- Lost teachable moments: A coach might see something in real time, but by the time the tech catches up, they're on a different swing or conversation.
- Session drag: Waiting around for uploads, spinners, or delayed results makes a 45-minute lesson feel like a 30-minute one.
Edge compute doesn't magically make the internet faster. It simply removes the network from the most time-sensitive parts of the loop, so your coaching rhythm stays intact.
Cloud-only workflows: a quick look at the bottlenecks
In a traditional cloud-only setup, a single swing might look like this:
- Record video on device.
- Upload full-resolution clip to the cloud.
- Run pose estimation and analysis in a centralized service.
- Save results, then stream overlays and feedback back down.
This can work well on fiber Wi-Fi in a quiet environment. But on a busy range, hotel Wi-Fi, or a cellular connection with dozens of other apps competing for data, that pipeline quickly turns into seconds (or tens of seconds) of delay.
The bottleneck isn't the model. It's the combination of:
- Upload time: even a few seconds per swing adds up over a session.
- Round-trip latency: time to send, process, and return results.
- Network variability: jitter, drops, or congested local networks.
Our edge-first approach restructures this pipeline so that the heaviest, most time-critical work happens before the network ever gets involved.
What we actually run at the edge
"Edge compute" sounds abstract until you decide exactly which tasks should live where. In BRD, we split responsibilities roughly like this:
On-device (coach / player phone)
- Frame capture and pre-processing: cropping, resizing, and normalizing frames so we're not shipping unnecessary pixels.
- On-device pose estimation (where supported): running a lightweight keypoint model directly on the phone for fast skeleton data.
- Swing detection and segmentation: identifying which frames actually contain a swing versus setup, waggle, or idle time.
- First-pass metrics: quick estimates like tempo, basic kinematics, and rough swing shapes to provide immediate visual feedback.
Near-edge / cloud
- Higher-precision models: heavier analyses that don't need to be instant, like deeper joint-by-joint breakdowns or comparative analytics across sessions.
- Long-term storage and rollups: keeping a history of swings, assignments, and progress so you can see trends over weeks and seasons.
- Team-level views: aggregations across players, squads, or entire programs.
The end result: the coach sees something meaningful almost immediately, and the "extra detail" can quietly stream in behind the scenes.
Real coaching scenarios where edge compute matters
On a crowded driving range
Think of a college practice: players on different mats, coaches walking up and down the line, and everyone swapping between range balls and on-course work. The network is shared, noisy, and not in your control.
With edge compute, you can capture a swing, see a traced swing shape, basic joint behavior, and initial AI observations in near-real-time—without depending on the range's router or your hotspot. The deeper metrics and storage can sync up once bandwidth is available.
Indoor facilities with patchy Wi-Fi
Not every hitting bay has enterprise-grade networking. Some are built in retrofitted warehouses, basements, or metal buildings that make signal coverage tricky.
An edge-first design means your core coaching loop—capture, review, draw, explain—still works smoothly even when the internet doesn't. When connectivity comes back, sessions sync without manual cleanup.
Travel lessons and on-course coaching
When you're walking 9 holes with a player or coaching at a tournament, you're often at the mercy of whatever network is available on your phone. Edge compute helps ensure that the quality of your session doesn't depend on the quality of the cell tower.
Design constraints we have to respect
Running models closer to the golfer isn't free. There are engineering and product tradeoffs that we keep front of mind:
- Battery and thermals: on-device inference can be demanding. We tune model sizes, frame rates, and sampling strategies so BRD feels responsive without cooking someone's phone.
- Device variability: not every coach or player has the latest flagship device. Our stack needs to degrade gracefully: lighter models and simpler overlays on low-end hardware, richer experiences where the silicon allows.
- Privacy by design: the more we can process locally, the less raw video we have to ship or store. That's better for latency and better for the player's privacy.
- Consistency of results: splitting work between edge and cloud means we have to be deliberate about model versions, calibration, and how we present numbers so they feel stable and trustworthy over time.
Edge compute isn't about pushing everything onto the phone. It's about carefully designing what belongs where so the coach's experience is as smooth as possible.
Where we're headed next
Edge compute is the foundation for a lot of what we want BRD to become. It's what makes two-phone stereo capture realistic, what makes real-time on-range overlays feel natural, and what allows programs to scale their tech without investing in expensive, proprietary hardware.
Over time, we'll keep pushing more intelligence closer to the golfer: smarter on-device classifiers, better swing-phase detection, and richer visualizations that update live as the player rehearses a change.
The north star is simple: when a coach pulls out BRD in the middle of a session, it should feel less like loading software and more like turning on an extra sense—fast, quietly powerful, and completely in sync with the way they already teach.