Tawori AI Processing Policy
Version: 2026-09-07-beta.1
Effective: 2026-09-07

This policy describes exactly when and how Tawori sends your content to
an AI provider. It applies to the free beta of Tawori.

1. What is processed, and only when you ask

Tawori never sends your content to an AI provider in the background.
An AI request happens only when you explicitly start one of these
operations, and each one requires your acknowledgement at the time of the
request:

  - CV analysis: the CV text you provided, the company name, role title,
    and job description you entered are sent to produce a directional
    comparison and a proposed tailored resume.
  - CV suggestions, reviews, and refinements: the relevant CV revision,
    job description, suggestions, your submitted evidence and decisions,
    and your refinement message are sent to generate or revise a draft.
  - PDF import: the PDF file you selected is uploaded and processed to
    extract its text.
  - Interview preparation: the linked resume revision, job description,
    and application context are sent to draft an interview preparation
    outline.
  - Answer feedback: your submitted answer, the interview question, and
    linked resume and application context are sent to produce feedback.
  - Interview practice: company and role details are used for question
    research; submitted design diagrams, notes, and the question are used
    for design assessment.
  - Saved job normalization: the job posting captured with the browser
    extension is processed to extract and organize job details.

2. Where your content goes

Requested operations are queued on the Tawori application server
(hosted on Railway) and sent through the OpenRouter API to pinned Google
Gemini models under Tawori's API credentials. Analysis and interview
preparation use google/gemini-3.8-flash. Suggestions, refinements,
extraction, normalization, answer feedback, and transcription use
google/gemini-3.5-flash-lite. OpenRouter may route a request between
compatible Google model providers for availability. Tawori requests zero-data-retention
endpoints and denies provider data collection. Your content is subject to
OpenRouter's terms and the terms of the routed model provider while it is
processed there. Tawori does not use your content to train models.

3. What is stored

The validated result of an operation is stored in your private Tawori
workspace so you can review it. For non-voice AI operations, Tawori also
stores the submitted input and context, the request sent to the AI service,
the original generated response, and processing details such as the model,
time, and errors. This includes submitted refinement messages, evidence,
and interview answers, and includes failed attempts and retries. These
records support troubleshooting and evaluation of AI quality; they are
not used to train models. Unsubmitted typing is not recorded in these
interaction records.

Interaction records are workspace-scoped and deleted with the workspace.
An enabled trace viewer and export are restricted to workspace owners.
Uploaded PDF originals are retained for source preview after a successful
import, until the source resume is deleted. Failed or expired imports are
scheduled for removal. You can request workspace deletion from Data &
Privacy.

4. Your responsibility to verify

AI output is a draft, not a fact. Tawori instructs the model not to
invent experience, skills, employers, credentials, or metrics, but you
must review and verify every claim before using generated content in a
real application or interview.

5. Changes

A new policy version replaces this document at a new URL or version
identifier, and Tawori will ask for your acknowledgement again before
the next AI operation. Consent receipts are bound to the exact policy
version and the exact inputs you acknowledged.

Contact: airaamane.pro@gmail.com