video1.0<iframe src="https://www.loom.com/embed/5caed2bab4ca4b3998251edcdcf84b53" frameborder="0" width="1662" height="1246" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12461662Loomhttps://www.loom.com12461662https://cdn.loom.com/os/loom-media-content/loom/60ddbe98-6b1b-40af-9857-69f0cbbca2e0/46a17a67-1d8d-4f0c-ae99-d5ca313218ce/sessions-thumbnails-5caed2bab4ca4b3998251edcdcf84b53-dff4a3763c856dc7.gif/binary?authToken=eyJraWQiOiJtaWNyb3MtZ3JvdXAvdGRwLW9zLWFyY2hldHlwZS9iOW5jNGxsMTMwYXUzYjduIiwiYWxnIjoiUlMyNTYifQ.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.XESlKvgyGSnRb1YQQ1eD8stOKkV4E05-8J3k7_YYA5LtQ9elSvUGGe53ggwqtscZSGPvSK6cQwRvayxmEiXBQKXyMcuqjgySYo0gCi_-mwhCSejEfZowTYRV4NCQR70875jh5FEOxCNX7N6uD1bNkNmaz4aHUwjjqtk_02d4IZjDhHLhA7CMWLaECXKLTdlW1tY6E74ZgNEVUv8ldoWkOvxtt9_PhNxzQfj-zd8nOx_fnmrGlQjOyqKOC_a92_vRmG_AlQnwyTzqsXmVV5hX_3oDgPi7qwgrSKDWMUfKiqv798w4qqb3DOe-Si-u6T_7z0NBiQm-N5Qm5rhFrI6kjA&Expires=1792293778&Signature=QirPUxc0ql1slH5J%7EWno2R1f7PqWHF%7Ej1Xsa%7EOd6AdTNB9EKXX3V4fel2KPEiPirpNxBpjRRmK0I-dg3pkmKMluqa1HbXNHepFB-ch2Vr6Kgv3xtiylFe%7ENMvx47lLMC--%7EmKtt9Un4YHjU42UHt1ExMkqPFw1vpLljBA4qBaG-2QpAOUsuJhs-d4zVXBBmDudfjRQaATNzNU7L23kRuZWfzoxBCcMuuTCh-InfXEmN%7ETSACQVGwOo9W1JqZyE2rTnGZ%7EInDJfjSFEMkB9dmXPqjNsE7U7eDq0cUES8gQKuni7EoDm8QHF9%7EvkEGEd9MyDadKfQ6LEq4mz8%7EpZgKlw__&Key-Pair-Id=KQOSYIR44AIC0356.397This Loom explains Sixth Sense, an AI product for FirstBus London to predict near misses and incidents involving London buses. Danish Asher and Dinesh describe how forward-facing bus camera data is processed with NVIDIA Cosmos, Corvive, and Vast to identify where and when risks are likely to occur, including pedestrian step-out scenarios, and to generate report cards and driver nudges. They say the goal at TfL is no one killed through London buses by 2030 and that the system supports driver coaching and safety reporting by showing risks on specific routes. Dinesh also shares that the backend uses Python and FastAPI, with Vast wrapping S3 object storage and Cosmos reasoning generating embeddings stored in a vector database, built with Cursor and Cloud Code.