‍

Shen AI SDK 3.0

SDK 3.0 is live today. This is the biggest release we've shipped. Not because of the feature count, but because of what it means for you and your users.

‍

Signal fusion (rPPG & rBCG)

Previously, the SDK analyzed both rBCG and rPPG signals in parallel but only chose the better one at the end. Analyze both, pick one, discard the other.

Now it works beat by beat. Both signals are processed in real-time, and the SDK knows the quality of each beat as it happens, for both rBCG and rPPG modality. Instead of picking a winner at the end, it selects the best beats from whichever source is strongest at any given moment.

Think of it as two sensors compositing the best data from each, moment by moment.

This means significantly better results across all skin tones. Signal fusion eliminates the performance gap that existed when relying on a single signal source.

‍

More accurate readings

We improved blood pressure accuracy by 20% in the elevated range (140–170 mmHg) and BMI by 45% thanks to a completely retrained body composition model trained on millions of real-world samples. 

‍

The SDK now tells you why you can trust a result

New Measurement Quality Guidance gives your users and your team real-time visibility into measurement conditions. Before the scan, users get coached on positioning and lighting. During the scan, they see live quality feedback they can act on. After the scan, results come with context, including a quality rating and, when conditions aren't optimal, a specific tip for improving next time.

For your engineering team: the Environmental Quality Index exposes 7 components via API: signal quality, camera noise, forehead coverage, light equality, and more. Use it for pre-scan optimization, post-scan filtering, or analytics dashboards.

For clinical use cases: Strict precision mode can now suppress Blood Pressure results entirely when measurement quality falls below threshold, preventing unreliable readings from reaching users. Off by default, configurable per integration.

‍

Face tracking that stays with you

The 3D face mask has been completely rebuilt. Previously, it was fitted once at the start of the scan and stayed static. If a user shifted position, tilted their head, or even adjusted their grip on the phone, the mask could slip and the measurement would fail.

Now, a neural network discovers the mask in a single step and realigns it periodically throughout the scan. A dual-tracker architecture runs in parallel: a main tracker keeps the measurement going while a refit tracker recalculates optimal positioning in the background. When the refit tracker finds a better fit, it transitions smoothly. No interruption, no restart. Even if someone runs their hand across their face, the mask snaps back into position and the measurement continues.

In practice: fewer failed scans from natural user movement.

‍

New user interface

The results screens for vital signs, pre-risk forms, and risk assessments have been completely redesigned. On by default, switchable if you prefer the previous look.

The shift? Every metric now sits on a visual scale. Users don't just see a number. They see where that number falls relative to the normal range. Understanding a result goes from "what does this mean?" to immediate clarity.

Blood pressure gets two presentation modes. Numeric values on a visual scale for detail-oriented users, or a simplified tile that highlights the range at a glance. You choose what fits your product.

‍

Wellness Score 2.0

Single measurements are noisy. Someone's heart rate at 3pm after coffee tells a different story than their weekly trend. Wellness Score 2.0 fixes this by averaging key metrics over a historical window using locally stored data.

The metrics it aggregates? Heart hate, HRV, breathing rate, blood pressure, stress index, cardiac workload, and BMI. Only valid measurements count. The score activates once two or more data points exist.

Same score, same scale, same interpretation. Just smarter inputs. The result now includes the time range it's based on (e.g., "Average of the last 7 days"), so users always know what they're looking at. 

‍

A guided measurement experience

SDK 3.0 ships with a completely new measurement interface, (you just need to turn it on in your playground). But the real change isn't visual. It's behavioral.

From the moment users enter the scan screen, they receive active positioning feedback: "Uncover your forehead." "Take off your glasses." "Turn toward a light source." "Illuminate your face more evenly." This guidance is situational. It responds to what the camera actually sees, not a generic checklist.

After the scan, the new results screen presents health data alongside Measurement Quality Guidance making results more actionable and easier to understand. This turns the SDK from a passive measurement tool into an active coaching experience. Users get better at scanning over time, which means better data quality with every use.

Your team controls the rules: configure which environmental conditions block the measurement start, which show warnings, and which stay silent, all through the Playground. Default behavior is preserved, so nothing changes unless you want it to.

‍

Note for integrations with blood pressure

You may notice measurement quality ratings are slightly more conservative than before. This reflects a new BP-specific quality index that is more clinically rigorous than the previous one. This is intentional. It gives your users and your team more honest quality feedback for blood pressure readings specifically.

‍

New language: Turkish

Turkish is now available as a supported language in the SDK. Enable it through your Playground configuration.

‍

‍

‍