MiCRAMultimodal content repurposing agent
Takes transcripts, articles, video, and audio and produces platform-specific outputs, with a consistency engine that checks generated text against every source.
- Year
- 2025
- Category
- AI / ML
I led a five-person QMIND team building MiCRA for Project X, a Toronto AI and data consulting firm. It takes long-form inputs (transcripts from YouTube, Zoom, or Whisper, articles, video, audio, and images) and turns them into platform-specific outputs like LinkedIn posts, emails, tweet threads, and meeting notes.
The interesting part is the consistency layer. Most "prompt in, content out" tools start hallucinating as soon as you give them several sources. MiCRA extracts entities across every input, aligns and normalizes terminology between them, and checks generated text against the sources before it leaves the pipeline. Generation runs on Gemini through a template-driven prompt builder, and a human-in-the-loop editor step closes the loop.