A connected insole
Continuous gait monitoring and real-time fall prediction, designed for elderly users in independent and assisted living.
Case study · Client anonymized — published with approval
Smart Insole — engineering a step before the fall
A connected insole, real-time AI and a defensible IP position — built end-to-end for a specialty healthcare provider.
Hardware cannot be vibe-coded.
AI-assisted code generation can accelerate firmware. It cannot design an antenna inside an insole, partition inference across device and cloud, close a sub-$50 BOM, or pass medical-grade qualification. Axon Labs operates where those decisions are made — system architecture, hardware design and commercial defensibility.
Project outcomes
- Production-ready device delivered end-to-end, from architecture through commercial sample.
- Defensible IP position established alongside the engineering deliverables, ready for patent filing.
- Real-time fall prediction — on-device and cloud — deployed in real users’ homes.
The problem
1 in 3 adults over 65 falls each year — and most never see it coming.
Falls are one of the most preventable causes of injury in elderly populations — and one of the hardest to monitor unobtrusively. Wristbands get removed, cameras feel invasive, and most consumer wearables track activity, not risk. A specialty healthcare provider came to Axon Labs with a different vision: predict falls before they happen, from a place no one ever takes off — the foot.
The client
A clear product vision, an unproven path.
No internal engineering path
Strong clinical understanding, but no team capable of taking the concept through hardware, AI integration, certification and patentability.
Build it — and defend it
Deliver the device end-to-end and secure the intellectual property to defend it commercially against fast-moving competitors.
The challenge
Five constraints, one product.
Every constraint was non-negotiable. Together, they defined the engineering envelope.
Tiny form factor
Sensors, MCU, antenna and battery — all inside an insole, without compromising LTE-M signal quality.
Sub-$50 BOM
Mass-production cost target leaves zero room for over-specified components or redesign loops.
Hybrid AI
On-device inference for fall prediction; cloud AI for trend analysis — partitioned for latency, privacy and power.
Always-on sensing
Continuous gait monitoring with no missed steps and minimal user intervention or charging.
Certification ready
Documentation, traceability and signal-integrity simulation — engineered for medical-grade qualification.
System architecture
From foot to forecast.
A four-layer architecture moves data from physical motion to predictive insight in under a second.
Pressure + IMU + GPS
Multi-sensor fusion captures gait pattern, balance and location.
MCU + on-device AI
Real-time fall-risk inference — runs even with no connectivity.
LTE-M, low power
Antenna co-designed for the constrained insole geometry.
Backend + analytics
Long-horizon trend modeling, alerts to caregivers and family.
End-to-end latency target: under 800 ms from sensor reading to actionable alert.
Engineering deep dives
Where the hard problems lived.
Antenna design, in a footprint that hates antennas
An insole is one of the worst RF environments in consumer electronics: surrounded by ground plane, sweat and a body that absorbs LTE-M signal. We co-designed a PIFA antenna with the PCB outline and enclosure, using full-wave electromagnetic simulation at every iteration, validated to LTE-M Cat-M1 on a 4-layer controlled-impedance stack-up.
Two AI brains, one coherent product
On-device: a quantized model detects fall-risk gait patterns in under 100 ms within an MCU memory and milliwatt power budget — offline by design. In the cloud: long-horizon models learn user-specific baselines, detect drift and generate caregiver alerts, improving through fleet-wide learning.
Commercial engineering
Engineering against $50.
In medical hardware, every dollar of BOM is fought for. The target was locked at architecture stage — every later trade-off was measured against it.
Right-sized silicon
Single-MCU architecture — no over-spec, no headroom we wouldn’t use.
Sensor fusion, not redundancy
Fewer sensors that work harder, validated through simulation.
Antenna co-design
Eliminated dedicated RF front-end real estate by integrating early.
DfM from day one
Manufacturing constraints were design inputs, not late-stage corrections.
IP & defensibility
Building the product and the moat around it.
Engineering services typically end at sample release. We treat patentable invention as a deliverable.
Identify novelty
We map the engineering decisions that are genuinely new — and separate them from prior art that already protects competitors.
Capture invention
Engineers maintain invention disclosures alongside design documentation, with timestamps and trace to architectural decisions.
Support filing
We work with the client’s IP counsel to articulate claims technically — translating engineering depth into defensible patent language.
Deliverables
What the client walked away with.
Named artifacts — every one of them auditable, reusable and ready for production handoff.
System Architecture Document
End-to-end architecture, design rationale and traceability from requirements to implementation.
Hardware Design Package
Schematics, layout, BOM and signal-integrity simulation reports.
Production File Set
Manufacturing-ready files: gerbers, assembly, test and DfM specifications.
Firmware Source Code
Full source, delivered under MIT license, with build, test and deployment documentation.
Technical & User Guides
Operational documentation, user-facing instructions and certification-aligned dossiers.
IP Disclosure Pack
Invention disclosures and technical claims documentation for patent filing support.
Discuss a confidential product program.
Tell us what must become technically and commercially real.