Camilo Herrera.Managua, Nicaragua
BACKEND · DATA · AI
Back to selected work
ElevenLabs at Una Mano / Applied AI

Reference conversations.
Human decisions.

This fictional text simulation shows how a reference’s account can be organized for a human reviewer. Compare a direct account with hearsay, and inspect the excerpts behind each field.

Fictional editorial reconstruction · Text simulation · No phone call
Reviewed October 14, 2026: one controlled team call reached a fictional record and the reviewer view was published. Contact returned to simulation mode; real reference use remains pending. This demo uses separate fictional conversations.

A conversation prepared for review

Fictional scenario · Human review required
Conversation transcriptSpanish-language agent
  1. AI agentT1

    Hola. Soy un asistente de inteligencia artificial de Una Mano. Esta conversación se transcribirá para que una persona revise su testimonio. Puede decidir no continuar. ¿Desea conversar sobre el trabajo del técnico?

  2. Fictional reference AT2

    Sí, puedo contar lo que vi.

  3. AI agentT3

    Cuénteme con sus propias palabras qué trabajo conoce y qué ocurrió.

  4. Fictional reference AT4

    Era un trabajo de plomería en mi casa. Yo estuve allí y vi una fuga en el lavamanos. El técnico cambió el grifo y, cuando terminó, dejó de gotear.

  5. AI agentT5

    ¿Recuerda cuándo ocurrió o si hubo alguna dificultad?

  6. Fictional reference AT6

    No recuerdo la fecha exacta. No vi otra dificultad durante esa visita.

  7. AI agentT7

    Gracias. Su relato quedará para revisión de una persona del equipo. Esta conversación no decide la aprobación del técnico.

Evidence for reviewReference A

Select a field to locate its supporting excerpt in the transcript.

Supporting excerpt · T4

“Yo estuve allí y vi una fuga en el lavamanos.”

Human decisionPending review

The account is evidence for a person to assess. A completed conversation does not approve a technician.

ContactAnswered
Conversation outcomeSuccessful
TestimonySpecific account
Human decisionPending
01 / Identify the interaction

The agent identifies itself as AI, explains transcription, and asks whether the reference wishes to continue.

How the technical flow works

The backend checks signed provider events and correlates them with a conversation already linked to an attempt and submitted version. Results retain the transcript, supporting excerpts, gaps, and contradictions. Duplicate delivery preserves the existing evidence, while storage failure permits redelivery.

Each excerpt must exist in a reference’s turn. This establishes the source of the quoted words. The reviewer still assesses the structured interpretation and the testimony itself. Contact, conversation outcome, testimony, and human approval remain separate.

Conceptual workflow. Backend checks use fictional events; one controlled team call was delivered separately. Real reference use remains pending.
Conceptual workflow. Backend checks use fictional events; one controlled team call was delivered separately. Real reference use remains pending.
Read the engineering case