Express interest

Carnegie Mellon University · Student Research Collective

[Machine Learning]

We work together across frontiers of modern ML — learning deeply, building carefully, and aiming to publish at the world’s top conferences.

Get involved Explore research

What we do

A collaborative lab for curious minds who want to ship ideas into papers.

We are a student-led research group at CMU. Members pick ambitious ML questions, run disciplined experiments, read the literature together, and push toward submissions at elite venues.

Large Language Models Interpretability Reinforcement Learning Multimodal Learning Representation Learning AI Safety & Alignment Computer Vision Efficient ML Reasoning & Agents Causal Inference Large Language Models Interpretability Reinforcement Learning Multimodal Learning Representation Learning AI Safety & Alignment Computer Vision Efficient ML Reasoning & Agents Causal Inference
01

Learn together

Paper clubs, shared notes, and rigorous critique — we treat reading as a craft.

02

Build experiments

From baselines to ablations — we design studies that can actually convince reviewers.

03

Publish upward

Our north star is clear: contribute knowledge at NeurIPS, ICML, ICLR, and peer venues.

Research agenda

Topics we explore

Interests evolve with the field. Right now we orbit questions around how models reason, how they fail, and how to make them trustworthy and useful at scale.

  • LLMs & reasoningChain-of-thought, faithfulness, agents
  • InterpretabilityMechanistic insight into trained systems
  • RL & decision-makingPolicy learning, reward design, control
  • Vision & multimodalityGrounded perception and generation
  • EfficiencyCompression, distillation, scalable training
  • Safety & alignmentRobustness, evaluation, responsible deployment

Target venues

NeurIPS ICML ICLR ACL EMNLP CVPR AAAI AISTATS

Join the collective

Interested in researching with us?

Drop your details — we’ll get an alert and follow up. No lengthy application.

Sends an email alert to the PI. First submit may ask you to confirm via FormSubmit.

Team

Who’s building this

ML Researchers 13

Shubhanandan T Y RL training dynamics, AI safety

Works on RL training dynamics and AI safety & alignment, building diagnostics that catch a failure mode before it shows up in the outcome metric: collapse inside GRPO-style fine-tuning loops, or a safety grader misjudging what it grades. Starting in the group with credit assignment in multi-turn LLM RL.

shubh442005 [at] gmail [dot] com
Naman Jain LLM jailbreaking, LLM efficiency

Comes to the group with research exposure in LLM jailbreaking and in improving LLM efficiency using the xlns logarithmic number system library.

namanjain9271 [at] gmail [dot] com
Kilian Shi Video segmentation, embodied AI

Comes to the group from research at the University of Notre Dame on sparsely supervised video segmentation with temporal memory. Now a first-year master’s student at CMU, currently working on a research project in the Xu Lab, with interests in agentic AI and embodied AI, including vision-language-action models.

haoyans [at] andrew [dot] cmu [dot] edu
Kobrin Goloveyko ML for embodied AI

Brings an interest in the ML side of embodied AI, especially systems where robots actively explore and collect data from their environment. Worked on a multimodal ASL translation project using computer vision and deep learning. Has raised edge efficiency in RL and active learning as possible directions, without settling on either or ruling out others.

kgolovey [at] andrew [dot] cmu [dot] edu
Shreya Kumari World models, closed-loop decision-making

Works on world models and closed-loop decision-making. Built on Archemist, an autonomous experimentation platform that ties experiment planning, measurement and optimization to ROS-based robotic control, and is interested in inference-time planning with learned world models.

23f2003785 [at] ds [dot] study [dot] iitm [dot] ac [dot] in
Prathin Avula Agentic AI, RAG pipelines

Works on agentic AI systems, building document and RAG processing pipelines on the cloud. Coauthored a paper on high-probability convergence of clipped SGD under generalized smoothness with heavy-tailed noise, which shapes an interest in making ML systems reliable.

avula.prathin1 [at] gmail [dot] com
Mohamed Zakkaria Efficient ML hardware, FPGA deployment

Works at the hardware end of efficient ML, with a bachelor’s thesis on deploying machine learning on FPGA published at IEEE VLSI-SoC 2025. Also interested in agentic AI and LLMs, and joins the group remotely.

mohamedzakkaria20 [at] gmail [dot] com
Amogh Hosamane LLM reasoning, multi-agent systems

Builds LLM-integrated learning systems and agentic AI workflows, with work at the National University of Singapore and on an AI-powered clinical simulation platform. Interested in LLM reasoning, multi-agent systems, RAG and efficient ML.

amoghvarsh9614 [at] gmail [dot] com
Anushka Paul Biomedical image analysis, multimodal AI

Brings an interest in biomedical image analysis and multimodal AI. Built an NLP pipeline using PyTorch, Sentence Transformers and FAISS for semantic retrieval over more than 1,500 research papers.

anushka011005 [at] gmail [dot] com
Mahmud Lawal Efficient ML for embodied AI

Interested in efficient ML for embodied AI, getting perception and navigation models to run well on CPU-only and edge hardware. Recent work includes a semantic object-finding robot in Gazebo with a CPU-only YOLO11n + OpenVINO detector, and an edge-AI surveillance system on a Raspberry Pi Zero 2 W. Wants to work on efficient vision models and vision-language-action models for robots, and joins the group remotely.

mahmudlawal75 [at] gmail [dot] com
Wajahat Ali Basharat Multimodal LLMs, multimodal retrieval

Works on multimodal large language models, with a focus on their architecture and multimodal retrieval pipeline. Starting in the group by leading a project on question-aware retrieval for knowledge-based visual question answering, so that different questions about the same image retrieve different evidence.

wajahatalibasharat073 [at] gmail [dot] com
Aniket Patil LLM reasoning, multi-agent systems

Interested in LLM reasoning, multi-agent systems and AI safety, especially making AI agents more reliable. A Computer Science and AI undergraduate at KLE Technological University and a research intern at IIIT Dharwad working on rPPG and multimodal EEG-rPPG analysis. Recently built a multi-agent meeting workflow engine with LangGraph, Gemini, RAG and MCP tools.

iamaniketpatil08 [at] gmail [dot] com
Utsav Jain LLM reasoning, coding agents

Interested in LLM reasoning and coding agents, especially how language models can use compiler feedback, static analysis and other tools to improve code generation and problem-solving. A student and software developer from India who builds real-world applications, and recently had his first pull request merged into Microsoft's Roslyn compiler.

utsavjain.com [at] gmail [dot] com