At codec.ai we are processing at a daily basis a large volume of input streams in different modalities: text, image, videos. Understanding and making sense of what this content is from a cultural point of view is a challenging task. Here, we will be presenting our multimodal search engine which makes possible to search text, image and video content.

We will be discussing traditional information retrieval approaches augmented with dense retrieval representations produced by neural networks (embeddings), dot product queries with Elasticsearch and approximate nearest neighbor techniques such as Locality-Sensitive Hashing (LSH) an Product Quantization (PQ).

The Search track is presented by OpenSource Connections