TECHNICAL PORTFOLIO // ENGINEERING & MODELING

agp-media: Unified Media Transcription, OCR & Qualitative NLP Toolkit

proprietary_sanitized

Modular command-line toolkit (CLI) built to process focus groups, in-depth interviews, and field survey data. Handles media conversion, audi...

[!NOTE] Sanitized Technical Specification (proprietary_sanitized): This technical sheet is an authorized public specification adhering to strict intellectual property compliance and Brazilian LGPD (Lei Geral de Proteção de Dados) compliance. Proprietary source code, production database instances, raw databases, confidential credentials, and client Personally Identifiable Information (PII) remain strictly protected and are not exposed.

Especificação Técnica Sanitizada (proprietary_sanitized): Ficha técnica pública autorizada em estrita conformidade com a Lei Geral de Proteção de Dados (LGPD) e diretrizes de propriedade intelectual. Código-fonte, instâncias de produção, credenciais confidenciais e dados pessoais identificáveis (PII) protegidos.

Ficha técnica pública autorizada sob diretriz proprietary_sanitized. Arquivos de áudio originais de entrevistas e formulários de pesquisa de campo permanecem protegidos sob estrito sigilo.

Organization / Context: Instituto de Pesquisas Aplicadas & Inteligência de Campo
Timeline: 2025 - 2026
Domain Category: Data Engineering
Technologies: Python, OpenAI Whisper, Google STT, Gemini API, PaddleOCR, Mistral, Pydantic, FFmpeg, CLI / Typer

Executive Overview

Modular command-line toolkit (CLI) built to process focus groups, in-depth interviews, and field survey data. Handles media conversion, audio transcription with multi-speaker diarization (Whisper, Gemini, Google STT), physical form OCR, and qualitative sentiment analysis into structured canonical JSON.

Architecture & Data Flow

The diagram below illustrates the sanitized high-level component topology and data processing pipeline:

flowchart TD
  A[Áudios & Vídeos Brutos de Campo] --> B[agp-media convert / compress / split]
  B --> C[agp-media transcribe: Whisper / Google STT / Gemini]
  D[Formulários Físicos / PDFs] --> E[agp-media ocr: PaddleOCR]
  C --> F[agp-media unify: JSON Canônico Estruturado]
  E --> F
  F --> G[agp-media analyze: Profiling Qualitativo & Sentimento]
  G --> H[agp-media export: Relatórios Executivos DOCX/MD]

Key Engineering Highlights

  • Multi-Engine Pipeline: Pluggable speech-to-text and LLM support for diarization, transcription, and sentiment analysis.
  • Multimodal Engineering: High-accuracy OCR for physical field survey sheets and unified structured JSON exports.
  • Operational Productivity: Packaged via pipx providing standardized command-line tooling for non-technical research analysts.