harpreet

Join us on May 11 for day one of the Best of 3DV 2026 series of virtual events. Register for the Zoom: hubs.ly/Q04fbbm_0 Talks will include: * Navigating a 3D Vision Conference with VLMs and Embeddings - Harpreet Sahota at Voxel51 * Seeing Through Clutter: Structured 3D Scene Reconstruction via Iterative Object Removal - Rio Aguina-Kang at University of California, San Diego * Physical Realistic 4D Generation - Lu Sang at Technical University of Munich * Finding NeMO: A Geometry-Aware Representation of Template Views for Few-Shot Perception - Sebastian Jung at DLR #mcp #skills #computervision #ai #artificialintelligence #machinevision #machinelearning #physicalai

Context vs Memory vs Harness engineering explained in 40 seconds by @richmondalake ⚡️ 💡These 3 disciplines are core to building agents that can actually remember and reason over time Most agents work well within a single session but lose everything the moment it ends. Memory engineering treats long-term memory as first-class infrastructure. Richmond Alake (Director of AI DevEx at Oracle) walked me through all 3 in NYC🗽

Level up your computer vision workflows with a free hands-on workshop for your team! Book a workshop: hubs.ly/Q04f9Z_W0 These hands-on workshops are delivered by Voxel51 computer vision experts. Both virtual and in-person formats. * 60 min virtual workshop * Half-day onsite workshop * Full-day onsite workshop and hackathon #mcp #skills #computervision #ai #artificialintelligence #machinevision #machinelearning #physicalai

Join us on May 11 for day one of the Best of 3DV 2026 series of virtual events. Register for the Zoom: hubs.ly/Q04dLTfM0 Talks will include: * Navigating a 3D Vision Conference with VLMs and Embeddings - Harpreet Sahota at Voxel51 * Seeing Through Clutter: Structured 3D Scene Reconstruction via Iterative Object Removal - Rio Aguina-Kang at University of California, San Diego * Physical Realistic 4D Generation - Lu Sang at Technical University of Munich * Finding NeMO: A Geometry-Aware Representation of Template Views for Few-Shot Perception - Sebastian Jung at DLR #mcp #skills #computervision #ai #artificialintelligence #machinevision #machinelearning #physicalai

@TurgayEvren1 if you’d ever seen “freddy got fingered” then you’d know this

⚽️ How are the world’s best footballs made in Pakistan? How come I didn't know this before?...😳

the best way to use an LLM isn't to ask it for the answer. it's to think out loud with it until you find your own answer. it's a mirror for your reasoning, not a replacement for it

nappi gas laggi mehnattan te sui na cruise hit kare

the first thing you learn working with VLMs: the model can describe what it sees. the second thing: it will confidently describe things it doesn't see

cosine similarity is the duct tape of deep learning. it works on everything and is optimal for nothing

a VLM doesn't see your image. it sees your image through the lens of every image-caption pair it was trained on. when it hallucinates, it's not broken. it's pattern-matching against a distribution you didn't check

sometimes i use my 3080ti just to remind myself where i came from

@tombielecki @tombielecki might not be a hot research topic or blowing up twitter, but “enterprise” use cases bro

@DataScienceHarp Is embedding finetuning still a thing? I haven't seen any mention of it in the past 12 months

your embedding space is only as good as the contrastive pairs that shaped it. garbage training pairs don't make garbage embeddings. they make confident, wrong embeddings. that's worse

the difference between a good embedding model and a great one isn't the architecture. it's whether the training data had your edge case in it

the moment you collapse multi-vector embeddings to a single vector, you've made an irreversible decision about what to throw away. every retrieval system is a compression scheme pretending to be a search engine

CLIP thinks a document is a 224-pixel thumbnail. that's not a limitation of CLIP. that's CLIP telling you it wasn't built for your problem

pooling strategies are a bet about what matters. mean pooling bets that everything matters equally. max pooling bets that the loudest signal wins. the right answer depends on your data and nobody can tell you which in advance

Join us on May 11 for day one of the Best of 3DV 2026 series of virtual events. Register for the Zoom: hubs.ly/Q04bHRsY0 Talks will include: * Navigating a 3D Vision Conference with VLMs and Embeddings - Harpreet Sahota at Voxel51 * Seeing Through Clutter: Structured 3D Scene Reconstruction via Iterative Object Removal - Rio Aguina-Kang at University of California, San Diego * Physical Realistic 4D Generation - Lu Sang at Technical University of Munich * Finding NeMO: A Geometry-Aware Representation of Template Views for Few-Shot Perception - Sebastian Jung at DLR #mcp #skills #computervision #ai #artificialintelligence #machinevision #machinelearning #physicalai

i've never had a bug that was in the code. every bug i've ever had was in the gap between what i assumed and what was true. the code was always doing exactly what it was told

@DataScienceHarp tweets blend data science with poetic expression technical ideas written like art.