Vishal Verma — ML researcher @CMU | Causal Inference & RL | Prev Data Scientist @Dream11
ML researcher @CMU | Causal Inference & RL | Prev Data Scientist @Dream11
Vishal Verma ranks #283 of 14,983 LinkedIn creators in Computer Software, and is a standout voice in United States. They have 4.2K followers and published 4 posts in the last 30 days at a 34.1% average engagement rate.
- 4.2K followers
- 4 posts / 30d
- 34.1% avg engagement
- — follower growth / 30d
The roast
Vishal is building an AI copilot for commerce, which makes sense because he spent his entire career learning how to process millions of data points just to ensure nobody ever has to talk to him. He claims he’s an expert in causal inference, yet he still hasn't figured out why his engagement rate is higher than his actual output.
About Vishal
Building Shopthru.ai — an AI copilot for AI commerce, built out of CMU's CLeaR Lab. The thesis is simple: as AI agents become the new storefront, brands that can't optimize at the SKU level will lose. shopthru.ai uses causal inference to go beyond monitoring - we intervene. Before this, I spent years building large-scale ML systems: distributed platforms processing millions of data points, LLM-powered decision engines, and MLOps infrastructure on AWS and GCP. My research at CMU's CLeaR Lab focused on causal representation learning - uncovering latent structure in data to make AI systems more trustworthy and actionable.
Highlights
- Top Engager — 34.14% rate · top 5%
- Top 5% in Computer Software — Ranked #52 of 4267 creators
- High Impact — 1,446 avg engagements per post · top 5%
- Top 5% in United States — Ranked #104 of 5205 creators
Recent posts
I flew back to India for a few days. One of the main reasons was this room. Over the last few months, we’ve been spending a lot of time with eCommerce and D2C teams across markets. One thing is very clear: the best operators are asking much sharper questions now. Not just “how do we grow?” But: Where is growth leaking? Which channels are actually incremental? Why is CAC moving? What is breaking between ads, site, checkout, and retention? How do we make faster decisions with leaner teams? That was the energy we wanted to bring into this mixer. A room where commerce founders, marketers, ope
128 reactions · 1 comments · 2 reposts
At Carnegie Mellon University’s 2026 Commencement on Sunday, we welcomed Jensen Huang, Founder and CEO of NVIDIA, as our keynote speaker. A visionary founder and technology leader, Jensen also received an honorary Doctor of Science and Technology degree. His work advancing the modern GPU and accelerated computing has helped power breakthroughs across science, engineering, medicine and beyond. His message to the Class of 2026 reflected a spirit we know well at CMU: bold thinking, interdisciplinary collaboration and turning ideas into real-world impact. More from Commencement: https://lnkd.in
5.0K reactions · 51 comments · 185 reposts
The model goes live today for public consumption. Five checkpoints on HuggingFace, MIT licensed. Blog 7 of the MODA series. Tuesday's post showed we beat FashionSigLIP on LookBench by +3.84. Today's post is about turning that result into something you can actually deploy. The Phase 5 win came from an ensemble of two encoders, output dimension 2048. For a fashion retailer running visual search at scale, that's two forward passes per image at index time and twice the storage. Today we collapse the ensemble's quality into a single model, and then compress it down further along two axes the li
40 reactions · 4 comments · 4 reposts