Every number below comes from live broadcast conditions — not a controlled test environment. We build for the hardest case first: real-time multilingual political media in minority languages, live, at scale.
Most AI captioning tools work well on a clean studio recording in English. Simourgh was tested on bilingual live streams, right-to-left scripts, dialects, and code-switching — the conditions where production systems actually fail.
That is why our numbers are meaningful: they were measured under the conditions your platform will actually face.
Whisper-based ASR with proprietary multi-vector indexing. Validated across hundreds of hours of continuous live production, including bilingual broadcasts and RTL script rendering.
Three complementary vector representations per segment with bounded reranking. Outperforms single-vector baseline (MRR 0.71) across live and archived content in multiple languages.
A proprietary policy for managing vector embeddings in live streams where transcripts evolve in real time. Solves a core problem that no dominant-language platform has addressed at production scale.
Correlates semantically related events across concurrent live broadcasts with a staged validation window. Validated across 3–5 simultaneous streams with inter-stream offsets of 4–25 seconds.
Phase 1 is live and in production. Each phase adds a new layer of understanding — from words, to images, to real-time reasoning.