Multimodal Input Pipeline
A unified processing pipeline for audio, video, images and documents
MiPi accepts heterogeneous input across multiple formats, preserves the original source and transforms it into a unified structured processing package. It handles documents, images, audio and video through format-specific pipelines while producing a consistent output model for downstream systems. Each package keeps a reference to the raw material together with extracted or transcribed text, processed content, processing artifacts and execution metadata. This preserves traceability from the normalized result back to the original source and allows later stages to work with a stable structure instead of format-specific details. MiPi is designed as an independent multimodal ingestion and normalization layer. Its role is to transform raw input into reliable structured data without embedding project interpretation or business logic into the processing pipeline.
- Role
- Creator, Product Architect, System Analyst
- System
- Java 21 / Spring Boot / Thymeleaf / PostgreSQL / MinIO / Speaches / LiteLLM / Docker
From raw input to structured data
Data Pipe is a multi-format data processing system for audio, video, images, text and documents. It preserves the original source, extracts or transcribes its content and produces a standardized package containing a reference to the raw data, the transcription, the processed text and execution metadata. The project is designed as a reusable ingestion layer that isolates downstream systems from source-specific formats and allows processing pipelines to evolve independently.