Santiago Cárdenas Jotty
Back to the cases

Multimodal AI · infrastructure

Roads AI Management

A workflow that turns Telegram text, voice, images, and documents into structured reports for infrastructure projects.

Role
Full Stack, AI and DevOps development
Context
Proyeckta · 2025

Visual evidence

Visual evidence01
Editorial diagram of the evidence, validation, and document-generation flow in Roads AI Management.
A sanitized system representation with no internal interfaces, records, or documents.

The challenge

Field teams create evidence in different formats and under changing conditions. The system needed to accept that material without introducing a complicated process and turn it into useful project information.

The response

I designed a staged workflow: Telegram intake, normalization, AI analysis, schema validation, and document composition. Each stage keeps enough context to diagnose failures without losing the source material.

Engineering judgment

AI was not treated as a black box. Extraction, interpretation, and presentation remain separate, allowing one stage to be retried without repeating the full process.

Professional case study shown without internal interfaces, data, or documents.

Key decisions

  1. Separate ingestion, analysis, document composition, and delivery.
  2. Preserve original evidence alongside processed results.
  3. Make asynchronous operations observable by design.

Verifiable outcomes

  1. Unified multimodal evidence in a single traceable workflow.
  2. DOCX and PDF generation with document previews.
  3. Explicit processing states, error handling, and retries.