TRACE 万析
EN中
Technology

Devices at the edge.Intelligence in the cloud.

TRACE splits the problem in three: hardware that touches the physical world, a record format that carries a scent between people and devices, and AI that learns from every confirmed version.

Architecture

Five layers,each with an honest status.

Dark blocks exist as prototypes. Amber blocks are in development. Outlined blocks are planned; dashed blocks are long-term vision. Data flows through planned blocks are not operational today.

Apps
Companion appPrototype built
Scent Card sharingPlanned
Partner APIVision
Cloud services
Formulation engine (AI)In development
Compatibility resolverPlanned
Record store & permissionsPlanned
Feedback learningPlanned
Scent record
Sensor fingerprint · conditions · description · blend · version · permissions · feedbackPlanned
Devices
AI NOSE — sensor array, sampling chamber, controllerExperimental
AI Diffuser — six dosing channels, MCU + Wi-FiPrototype built
Physical world
Real samples · aroma ingredient bottles · rooms and peoplePhysical
How it fits together

From readingto recipe.

  • 01

    What AI NOSE measures

    The response pattern of a gas-sensor array to a sample in a closed chamber, plus time, temperature, humidity, a blank and a reference reading.

  • 02

    How a scent is represented

    As a record that keeps the sensor fingerprint, the conditions, the person’s description and — once created — a blend: proportions across the six ingredients of a compatible device.

  • 03

    What the AI contributes

    It maps language, samples and feedback to candidate blends within the ingredients a device actually holds, and learns from each version people confirm. Public perfume catalogues and olfaction research inform candidate search.

  • 04

    How the device plays it

    The AI Diffuser doses its six ingredients in the stored proportions. A compatibility check runs first: if an ingredient is missing, it says so.

  • 05

    How it improves

    Feedback after each playback — closer, stronger, less sweet — is attached to the record and becomes the brief for the next version.

Validation

Can a scent bereproduced? We will measure it.

Pass thresholds are set before each test. Accuracy and reproduction quality have not yet been measured, and TRACE will not publish performance claims until they are.

01

Baseline

Measure a controlled reference sample and a blank background. Play the same version repeatedly with fixed settings.

Measured: Deviation between reference and output

02

Correction

AI NOSE samples what the diffuser actually releases; the system adjusts within safe limits and plays again.

Measured: Compared with fixed playback under identical conditions

03

Retest

Repeat across days, batches and devices. Blind panellists rate similarity and intensity in random order.

Measured: Deviation, blind similarity score, time taken

Open questions

What still hasto be proven.

  • ?

    Ingredient coverage

    Which reference scents can six ingredients approximate well enough?

  • ?

    Sensor stability

    How stable are readings across devices, batches and days?

  • ?

    The room

    How much do room size, airflow and residue change what people perceive?

  • ?

    Safe release

    Which release rates stay within indoor exposure limits?

Principles

Safety first

IFRA-compliant aroma materials, indoor exposure limits and electrical and radio certification are prerequisites for any product release.

Private by default

Formulas and raw sensor data stay private. Sharing grants playback rights only. Training on personal data is opt-in.

Status before claims

Every capability is labelled by status. Demonstrations at events use preset scents and say so.

See the end-to-end flow