Centralized headless knowledge curation, multi-format audio transcription, PostgreSQL pgvector dense semantic search, and state-gated GraphRAG retrieval service for the Dinsor enterprise platform.
Monitors databank intake, sweeps staged files, extracts structured text from Office presentations (PPTX), Word (DOCX), Excel (XLSX), and PDFs into dense vector embeddings.
Transcribes spoken recordings (.mp3, .wav, .m4a) using multimodal AI models. Generates companion executive summaries and action decision markdown (.transcript.md).
Discovers named entities (concepts, systems, people, technologies) and maps relational triples to construct persistent, navigable knowledge graphs.
Shields uncurated data from search queries. Synthesizes grounded answers fusing vector similarity and knowledge graph context with verified source provenance citations.
Permanent file vaulting with digital signatures (.curated). Features on-the-fly copy watermarking for PDFs, images, and documents without altering stored master assets.
Centralized polling engine communicating with external client platforms (e.g. DinsorOffice) via scheduled intake drain loops with atomic checkouts and completion ACKs.