CV
Ph.D. candidate at CMU Robotics Institute. On the job market for research scientist roles (graduating late 2026). Download the PDF for the latest version.
Contact Information
| Name | Gautam Rajendrakumar Gare |
| Professional Title | Ph.D. Candidate, Robotics Institute, Carnegie Mellon University |
| ggare@andrew.cmu.edu |
Professional Summary
I make frozen vision-language models do new things without touching their weights, treating adaptation as an inference-time problem: optimizing prompts (DetPO, ECCV 2026), selecting image-conditioned attributes, and reading activations (ACL Findings 2026). I study when and why multimodal models generalize (ICML 2024 Oral) and prove these methods in medical imaging (5 ISBI papers, 4 U.S. patents). On the job market for research scientist roles (graduating late 2026).
Education
-
2021 - 2026 (expected) Pittsburgh, PA
Ph.D. (candidate)
Carnegie Mellon University
Robotics
- Advisors: Prof. John Galeotti and Prof. Deva Ramanan.
- Focus: inference-time adaptation and alignment of vision-language models; generalization; medical imaging AI.
-
2019 - 2020 Pittsburgh, PA
M.S.
Carnegie Mellon University
Electrical and Computer Engineering
- Courses: SLAM, Computer Vision, Machine Learning, Optimization, Deep Learning.
-
2012 - 2016 Bangalore, India
B.E.
B.M.S. College of Engineering
Electronics and Communication
Research Experience
- Soft-Prompt Tuning of Vision-Language Models for Few-Shot Object Detection · Robotics Institute, CMU · Nov 2025 – Present. Designed a soft-prompting framework that learns continuous prompt embeddings via gradient-based optimization, enabling task adaptation without modifying VLM weights. Outperformed LoRA fine-tuning in the 10-shot setting, showing lightweight prompt optimization can match or exceed parameter-efficient fine-tuning at low data regimes.
- Iterative Prompt Optimization for Few-Shot Object Detection (DetPO) · Robotics Institute, CMU · Sep 2025 – Jan 2026. Developed an iterative prompt-optimization framework using false-positive and false-negative feedback to improve few-shot detection. Achieved state-of-the-art results with contemporary VLMs (e.g., Qwen3, Gemini3) on the RF20 dataset, outperforming specialist models such as GroundingDINO in a black-box setting.
- Causal Discovery for Classification · Robotics Institute, CMU · Jan 2023 – Aug 2025. Designed a Markov-blanket-based causal discovery algorithm for feature selection. Demonstrated improved accuracy and interpretability on medical datasets.
- Generic Human-Interpretable Lung Ultrasound Biomarkers · Biomedical Image Guidance Lab, CMU · 2021 – 2025. Developed interpretable lung ultrasound biomarkers enabling causal and contrastive learning for diagnosis. Showed biomarker-based features match deep features in downstream performance.
Professional Experience
- AI Research Intern · Robert Bosch LLC (Bosch Research), Pittsburgh, PA · May 2026 – Aug 2026. Researched grounding of vision-language-action (VLA) models for autonomous-vehicle applications, connecting language-conditioned action reasoning to visual scene evidence.
- Computer Vision Research Intern · DawnLight Inc., Palo Alto, CA · Jun 2021 – Aug 2021. Developed efficient multimodal action recognition networks (RGB + IR) for clinical monitoring.
- Senior Data Scientist & Software Engineer · Sling Media R&D (Dish Networks), Bangalore, India · Jul 2016 – Jul 2019. Led analytics pipeline for user churn modeling and AirTV Player Android app interface optimization. Improved image caching, reducing app startup time by 20%.
- Summer Research Fellow · IIEST, Shibpur, India · Jun 2015 – Jul 2015. Built a real-time tele-operated robotic arm system with RGB-D feedback.
Patents
- US12154354B2: System and Method for Labeling Ultrasound Data (granted 2024)
- US20240177000A1: System and Method for Training Machine-Learning Models with Probabilistic Confidence Labels (2024)
- US20240177445A1: System and Method for Direct Diagnostic and Prognostic Semantic Segmentation of Images (2024)
- US20240249498A1: System, Method, and Computer Program Product for Extracting Features from Imaging Biomarkers with ML Models (2024)
- US10547891B2: Apparatuses, Systems, and Methods for Adding Functionalities to Control Buttons on a Remote Control Device (granted 2020)
- IN201641006819: An Improved Faucet with Automatic Flow Control Using Proximity Sensor (2016)
- IN201641036243: Add-on Smart Device for Charging Portable Devices (2016)
Fellowships and Awards
- Liang Zhao Endowed Fellowship, CMU (2025)
- CMLH Translational Fellowship, CMU (2022)
- NSF XSEDE Compute Grant, 130,500 GPU hours (2020)
- Neocortex EUP Compute Award, NSF PSC (2020)
Teaching and Service
- Guest Lecturer, 16-720 Computer Vision, CMU (Fall 2026)
- Research Mentor, prompt-echoing VLM study (Ask Twice, Look Twice, arXiv 2026)
- Research Mentor, Prognostic Value of Lung Ultrasound Biomarkers for CHF Readmission Risk (arXiv 2026)
- Teaching Assistant, 16-720 Computer Vision, CMU (Spring 2022, Fall 2021)
- Reviewer, ECCV (2026), eXCV Workshop at ECCV (2026), MICCAI (2022-2024), HSMR (2023)
- Invited Panelist, LL-COVID19 MICCAI Workshop (2021)
- Mentor, CMU AI Mentoring Program (2021)
Media Coverage
- “CMU and others propose a highly practical ‘model generalization’ metric, improving out-of-distribution accuracy by 6% | ICML 2024 Oral”, New Zhiyuan (Dec 2024), covering LCA-on-the-Line. article
- “Seeing Beneath The Surface”, LINK Magazine, CMU School of Computer Science (Summer 2022)