At GagaMuller (Dublin, remote) I built ChatTuesday, a multimodal RAG chat system that combines LLMs with text-to-image and text-to-video diffusion models. The API is built with FastAPI and uses pgvector for semantic search over text and media.
Results
- 30% reduction in end-to-end generation latency through caching and batching.
- 25% improvement in cross-modal coherence (text vs. generated image/video) via better retrieval and prompt design.
The stack serves as a single interface for conversational search and multimodal generation, with embeddings and metadata stored in PostgreSQL for fast retrieval.