PiriZero · Olfactory Intelligence

Calibration without surveillance.

How we handle data. What we hold. What we never touch.

The engine needs one point in time. It does not need your history.

PiriZero produces forward-projecting olfactory predictions from a single assessment point. Four inputs enter the engine. Each is processed into physics. None of the raw inputs are stored.

No model. No machine learning. No inference that accumulates over time. The engine runs on precomputed physics laws — deterministic equations derived from receptor biology, hormonal science, and skin chemistry research. A language model puts words to the numbers the engine produces. It decides nothing. It computes nothing. The physics is done before the LLM sees the output.

This is Privacy by Design — not a compliance layer added after the fact, but a structural consequence of how the system works. A deterministic physics engine that forward-projects from a single point has no architectural need for longitudinal data. We are not choosing to discard your inputs. We are discarding them because the engine has already extracted everything it needs and the raw data has no further function.

Four Inputs. Four Discards.
Input 01 · Constitutional Map
Encoded passport. Physics extracted. Passport discarded.
The CM passport carries encoded constitutional data — a compressed representation of receptor phenotype and sensory architecture. The engine decodes it into physics parameters. The passport itself is not stored. What persists is the computed vector, not the source document.
Input 02 · Thermal Capture
Image processed. Thermoregulation phenotype extracted. Image discarded.
A thermal image is captured at the moment of assessment. The engine extracts thermoregulation phenotype — how this person's autonomic system responds to temperature — and maps it into the receptor model. The image is never stored. It is processed in session and destroyed.
Input 03 · Voice Analysis
Recording processed. Biomarkers extracted. Recording discarded.
A voice sample is analysed for stress state, autonomic tone, and neural activation pattern. These biomarkers are translated into modifier values that adjust the olfactory prediction. The recording is not stored. It does not leave the session.
Input 04 · Questionnaire
Answers processed. Receptor preferences extracted. Answers discarded.
Questions are designed to infer receptor biology without declaring it — the subject does not know what is being measured. Once the engine has extracted receptor sensitivity scores and hormonal phase calibration, the answers themselves are discarded. The questions are an instrument, not a record. In the pilot phase, two calibration points are gathered per session; same logic applies.
Data Held
01
What We Hold
  • Email address
  • Your olfactory profile — six computed dimensions
  • Fragrance recommendations generated for your session
  • Phase-specific predictions (sensitivity, concentration, longevity)
  • Current hormonal state declaration — one field, no history
02
What We Do Not Hold
  • Assessment question answers — discarded after computation
  • Voice recordings — processed in session, not stored
  • Thermal images — processed in session, not stored
  • Cycle dates, period history, pregnancy data
  • Location, age, demographics
  • Browsing behaviour or analytics
  • Data sold or shared with third parties
03
Hormonal Calibration

The engine calibrates across your hormonal cycle. To do this, it asks you to declare one current state — cycling, follicular, luteal, pregnant, postpartum, menopausal. We store that single declaration. We do not store when you made it, how it has changed, or any history of previous states. You update it when you choose. We never infer or track anything about your body beyond what you explicitly tell us in that moment.

Privacy by Architecture
Structurally incapable
Not a policy. A consequence.
Most privacy policies describe what a system chose not to do with data it collected. Ours describes what the engine has no mechanism to do. Forward projection from a single assessment point does not require historical data. We cannot build a longitudinal profile of you because the engine does not need one and was never designed to accept one.
Physics engine, not AI model
Precomputed laws. No learned inference.
The engine runs on deterministic equations — precomputed from receptor biology, hormonal science, and skin chemistry research. There is no model that learns from your data. The LLM is a copywriter: it receives numbers from the engine and produces language. It decides nothing about your profile. A human copywriter with the same numbers would produce the same report.
No surveillance loop
The engine does not watch you.
The engine does not improve by observing your behaviour over time. It does not log what you wear, infer new states from patterns, or refine its model by tracking you. Your profile is fixed at assessment. If your biology changes — pregnancy, menopause, significant health shift — you recalibrate by choice. Not because we noticed.
Receptor, not demographic
Biology measured directly.
PiriZero does not use age, ethnicity, or population averages as proxies for receptor sensitivity. It infers receptor biology directly from response patterns across the four input channels. Demographic data is not only unnecessary — it would introduce noise into a system designed to measure individuals, not populations.
GDPR
European data standards, met by design.
Data held is the minimum necessary to deliver your report and enable voluntary recalibration. No secondary processing. No profiling beyond olfactory prediction. Subject access, correction, and deletion requests handled within seven days. Deletion is complete — email, profile, all associated records removed.
Pilot phase
Research data handled identically.
During the current pilot study, session observations are logged for validation — fragrance responses, perception gap data, preference rankings. This data is anonymised at collection. It validates engine predictions; it does not profile individuals. Pilot participants are informed at enrolment.
Your Rights
04
Access & Correction
  • Request a copy of your profile at any time
  • Correct your hormonal state declaration at any time
  • Request full account deletion — no conditions, no waiting period
  • Response within seven days
05
Contact

All data requests are handled directly. No automated systems. A person reads and responds.

theashzero@gmail.com

AshZero Ltd · Paris

PiriZero is built by AshZero. The privacy architecture described here is not a policy layer added after the fact — it is a consequence of how the engine works. A deterministic physics engine that forward-projects from a single point has no architectural need for longitudinal data. We did not choose minimalism as a principle. We arrived at it from the physics.

A predictive model may be introduced in future — trained on physics outputs from the validation dataset, never on raw biometric inputs. The privacy architecture does not change. A model that learns from numbers derived from discarded signals is not a surveillance model. It is an extension of the same principle: compute from biology, store only physics.

The engine does not need your history.
It only needs to know where you are now.