Fahaam Qureshi

Meta Ai Hackathon 1M 2026 @cerebralvalley @meta Twitter: https://x.com/FahaamQureshi Github: https://github.com/fahaamqureshi/BLINDSPOT_CI BLINDSPOT CI helps developers evaluate vision models using human reference labels, Python scoring and evidence they can inspect. This complete demo combines the product tour with a recorded live workflow: initialize a workspace, import fresh photos, run the evaluation, inspect a missed object, save a review note and export the results. It also shows CLI and project-skill integration for coding agents. CHAPTERS 0:00 Product tour and use cases 3:00 Next: live demo 3:04 Fresh-data live demo: terminal to web 4:08 Run the real evaluation 4:28 Read results and coverage 4:46 Inspect the missed object 5:08 Review and export 5:26 Coding-agent integration The second study tested six previously unused COCO photos: five passed and one distant sign was missed against the original human references. The agent selected five cases; a scripted completion reused five predictions and made one additional SAM call. Decision: NEEDS_REVIEW. Twelve logical requests total, estimated USD 0.033. This is a small selected study, not a benchmark or safety approval. Practice footage is labeled. Real results are preserved recordings. Waiting is shortened; enlarged crops are editorial. Terminal footage starts after installation and displays actual command output. No ongoing livestream is implied. SAM 3.1 supplies vision predictions; Muse Spark chooses bounded tests; Python computes scores. Skill installation and JSON workflows were tested; a separate Claude Code session is not shown. Project: https://github.com/fahaamqureshi/BLINDSPOT_CI CREDITS Music: Happy Beats / Business Moves Vol. 12 by Sascha Ende, https://ende.app/en . CC BY 4.0: https://creativecommons.org/licenses/by/4.0/ . Edited volume and fades. Photos: source-listed CC BY 2.0, https://creativecommons.org/licenses/by/2.0/ . Adaptations: scaling, cropping, reference and prediction overlays. Original annotations: Open Images and COCO. Photo creators and original sources: Floris Looijesteijn: https://www.flickr.com/photos/florisje/7508445968/ agirlnamednee: https://www.flickr.com/photos/28072695@N03/16196232804 Nick Hubbard: https://www.flickr.com/photos/nickhubbard/3338181218 Victoria Padevit Brown: https://www.flickr.com/photos/photolady2000/1867151184 John Trainor: https://www.flickr.com/photos/trainor/5087211155 Rinaldo Wurglitsch: https://www.flickr.com/photos/wurglitsch/2071721182 SKR_RGR: https://www.flickr.com/photos/94444025@N07/19312058842 teteria sonnna: https://www.flickr.com/photos/sonnnateteria/19638945535 Sean Loyless: https://www.flickr.com/photos/haggismac/3274992844 Dane Low: https://www.flickr.com/photos/danelow/2455933232 Josbert Lonnee: https://www.flickr.com/photos/98552965@N05/9539412995 Carter Brown: https://www.flickr.com/photos/paintballphotos/14699039780 Paul Kimo McGregor: https://www.flickr.com/photos/10677920@N05/3838679592 Benoit Brouillette: https://www.flickr.com/photos/benoit59/9379650499 Shane Lewis: https://www.flickr.com/photos/presencephotography/2746974761 Andrew Melnychuk: https://www.flickr.com/photos/am_8/16311147631 John W. Schulze: https://www.flickr.com/photos/gruenemann/2050422645 JamesZ_Flickr: https://www.flickr.com/photos/jamesz_flickr/4341215320 Henks Pics: https://www.flickr.com/photos/henkspics/12302168354 shalbs: https://www.flickr.com/photos/shalbs/3351207312 Yoshikazu TAKADA: https://flic.kr/p/e5z7C6 muneaki: https://flic.kr/p/8uDR4W ell brown: https://flic.kr/p/9TMHoX piddix: https://flic.kr/p/6mScoN Open Images: https://storage.googleapis.com/openimages/web/factsfigures_v7.html COCO: https://cocodataset.org/#download Video: Richard L, https://www.pexels.com/video/33049294/ ; George Morina, https://www.pexels.com/video/5687586/ . https://www.pexels.com/license/ . Edited clips; video examples lack accuracy labels. Speech reference: LibriSpeech, Panayotov, Chen, Povey and Khudanpur, https://www.openslr.org/12 , CC BY 4.0. The film displays a saved transcript. Creator: https://x.com/FahaamQureshi