Kartik Sharma

I’m an incoming MSR student at Carnegie Mellon University (Fall ’25) focused on building intelligent systems that learn and reason across modalities. Previously, I was a Senior Engineer at Samsung R&D Bangalore, where I built safety filters for Samsung’s LLMs/LVMs and shipped on-device models to flagship devices. I earned a dual degree in Computer Science & Economics from BITS Pilani, completing research with Prof. R. Venkatesh Babu (Long-tail Image Generation with StyleGANs) and Prof. Prof. Poonam Goyal (Multimodal Crop-yield Prediction). I now aim to advance RL + Multimodal reasoning for robotics with safe, transparent, real-world deployment.

Email  /  CV  /  Github  /  Scholar  /  LinkedIn

Research

My research interests are broadly at the intersection of vision–language, robotics, and reinforcement learning. I am fascinated by how humans integrate multiple sensory signals—sight, sound, and language—to reason and act in the world, and I aim to replicate this ability in artificial systems. I study multimodal learning and reasoning, focusing on fusing vision, language, and action to enable robust decision-making in dynamic environments. My prior work spans language–vision reasoning, domain adaptation under distribution shift, and robustness in long-tailed settings, leveraging tools such as contrastive learning for cross-lingual grounding, generative models (GANs/diffusion) for limited-data scenarios, and retrieval-augmented pipelines for LLMs. Going forward, I am particularly interested in exploring methods to incorporate human priors by learning transferable skill representations across humans and robots.

I am open to exploring new research directions as well, so please feel free to reach out!

Publications

Experience

Research Engineer — Samsung R&D Institute India (Nov 2023 – May 2025)
Developed a multilingual safety filter for Samsung’s LLM with 95% accuracy across 12+ locales, deployed on flagship devices with 45% smaller models. Improved cross-lingual image grounding for better object localization in low-resource languages.

Data Scientist — PrivateBlok (Feb 2023 – Nov 2023)
Built PrivateBlok's MVP, a GPT-3.5-powered financial QA chatbot for 10K+ companies. Enhanced retrieval accuracy with a custom re-ranking algorithm and created a temporal knowledge graph for detailed financial insights.

Project Assistant — Video Analytics Lab, IISc Bangalore (Aug 2022 – Jan 2023)
Improved long-tailed image generation using StyleGANs, achieving 19% better FID scores. Published work at CVPR 2023, setting a new state-of-the-art for long-tailed datasets like ImageNet-LT, CIFAR10-LT, and iNaturalist-LT.

Software R&D Intern — Samsung R&D Institute India (May 2022 – Jul 2022)
Developed an object-action detection system with 82% precision and optimized transformers for large-scale action recognition.

Independent Projects

Constructed a speech dataset from synthesized data and implemented a trigger word detection model with over 90% accuracy. Trained a GRU(Gated Recurrent Units) to detect when someone has finished saying the word "activate".

Implemented real-time object detection on a car dataset using the YOLO model, which was further improved using a U-net architecture. The YOLO model was stacked with Non-max suppression layers using IOU grid analysis to obtain the most accurate boundary boxes.

Used transfer learning on the VGG-19 network to generate new artistic images. Implemented a cost function that minimizes the content and style cost by running both the images through the pre-trained VGG-19 model.

Thank you for visiting my site, and if you have any inputs, I would love to hear from you! Feel free to steal this website's source code. Do not scrape the HTML from this page itself, as it includes analytics tags that you do not want on your own website — use the github code instead.