generated: '2026-08-14'
method: derived
source: >-
  src/sensestreet/sensestreet_client.py (SenseStreetClient.dict2conv) and
  src/sensestreet/anonymise.py in the first-party PyPI package `sensestreet` 1.0.16
note: >-
  Sense Street publishes no OpenAPI and no schema document, so there is no $ref
  graph to walk. This entity model is read directly off the conversation payload
  the company's own Python client constructs and uploads — every field below
  appears verbatim in the shipped source. Field types are stated only where the
  source makes them unambiguous. Nothing is invented; where the client does not
  reveal a field, it is absent here rather than guessed.
entities:
- name: Conversation
  description: >-
    The top-level unit submitted for extraction. A chat between a bank and a
    client, carrying its participants and its ordered messages.
  id_field: chat_id
  fields:
  - {name: chat_id, type: string, note: 'client-assigned; the SDK helper mints "<ISO timestamp>Z_auto"'}
  - {name: participants, type: array<Participant>}
  - {name: msgs, type: array<Message>}
  submitted_via: [POST /api/v1/predict_rfqs, POST /api/v1/conversations]
- name: ChatSnippet
  description: >-
    A cut-out of a live chat rather than a full conversation in Sense Street's
    own format. Submitted to a separate operation because its shape differs from
    Conversation; the README states this explicitly.
  submitted_via: [POST /api/v1/predict_chat_snippet_rfqs]
- name: Participant
  description: One side of a conversation — a named person at a bank or a client firm.
  fields:
  - {name: login_name, type: string, note: 'chat platform login; the join key used by Message.login_name'}
  - {name: role, type: string, enum: [Bank, Client]}
  - {name: company_name, type: string}
  - {name: first_name, type: string}
  - {name: last_name, type: string}
- name: Message
  description: A single chat message inside a conversation.
  id_field: msg_id
  fields:
  - {name: msg_id, type: string}
  - {name: content, type: string}
  - {name: message_time_utc, type: string, format: iso8601}
  - {name: login_name, type: string, note: 'references Participant.login_name'}
- name: RFQ
  description: >-
    The extracted request-for-quote — the product Sense Street sells. Returned by
    the predict operations and written into the processed conversation file. Its
    field shape is not publicly published; it is only visible to authenticated
    callers, so no fields are listed here rather than guessed.
  fields: []
  extraction_target_default: RFQ_CREDIT
  note: >-
    The `target` query parameter on the file status/download operations selects
    the extraction target. RFQ_CREDIT is the SDK default; other targets exist for
    other asset classes (the product markets Fixed Income and Commodities) but
    their identifiers are not published.
- name: BatchFile
  description: >-
    A server-side record created when conversation files are uploaded. The unit
    the asynchronous batch workflow is tracked by.
  id_field: file_id
  fields:
  - {name: file_id, type: string, note: 'server-issued, one per uploaded file'}
  - {name: status, type: string, note: 'shape not published; returned by GET /conversations/status/{file_id}'}
  - {name: uploaded_at, type: timestamp, note: 'surfaced as the value in the history map'}
  operations:
    create: POST /api/v1/conversations
    status: GET /api/v1/conversations/status/{file_id}
    download: GET /api/v1/conversations/file/{file_id}
    history: GET /api/v1/conversations/history
- name: BondReferenceData
  description: >-
    Reference data for bonds, uploaded so extraction can resolve instruments
    mentioned in chat. Uploaded as a file; the record shape is not published.
  operations:
    create: POST /api/v1/bonds
- name: ERFQ
  description: >-
    Electronic RFQ records uploaded for reconciliation against the RFQs extracted
    from conversations. Uploaded as a file with file_type=ERFQ.
  operations:
    create: POST /api/v1/upload_file
- name: BloombergChatXML
  description: >-
    Bloomberg chat export XML — the raw source format the anonymisation helper
    operates on before upload. Its identity-bearing fields are enumerated because
    the SDK names each one it scrubs.
  fields:
  - {name: LoginName, pii: true, anonymisation: hashed}
  - {name: FirstName, pii: true, anonymisation: pseudonym}
  - {name: LastName, pii: true, anonymisation: pseudonym}
  - {name: UUID, pii: true, anonymisation: blanked}
  - {name: FirmNumber, pii: true, anonymisation: blanked}
  - {name: AccountNumber, pii: true, anonymisation: blanked}
  - {name: EmailAddress, pii: true, anonymisation: derived-from-LoginName}
  - {name: CorporateEmailAddress, pii: true, anonymisation: derived-from-LoginName}
  - {name: CompanyName, pii: true, anonymisation: 'only when bank_pattern is supplied'}
relationships:
- {from: Conversation, to: Participant, type: has_many, via: participants}
- {from: Conversation, to: Message, type: has_many, via: msgs}
- {from: Message, to: Participant, type: belongs_to, via: login_name}
- {from: Conversation, to: RFQ, type: has_many, via: extraction}
- {from: BatchFile, to: Conversation, type: has_many, via: uploaded file contents}
- {from: RFQ, to: BondReferenceData, type: belongs_to, via: instrument resolution}
- {from: RFQ, to: ERFQ, type: has_one, via: reconciliation}
- {from: BloombergChatXML, to: Conversation, type: has_many, via: parsed source export}
render: null