Open to opportunities

Pranamya Deshpande

@pranamyadeshpande

Message

AI/ML research intern focused on LLM evaluation, agent robustness, and explainable deep learning.

India

Message

What I'm looking for

I’m looking for a research-driven AI/ML role where I can evaluate and stress-test LLM agents, build explainable deep learning systems, and contribute to trustworthy AI. I value strong mentorship, reproducible experiments, and real-world deployment thinking.

I’m an AI/ML research intern working at the intersection of LLM evaluation and interpretability. My recent first-author research focuses on tool-using LLM agent robustness (AgentMorph) and on understanding where and how small language models localize, compress, and edit knowledge.

With AgentMorph, I introduced trajectory-level metamorphic testing for tool-using LLM agents, designing meaning-preserving mutations and deterministic benchmarks with rule-specific equivalence checkers. I also evaluated multiple open-weight instruction models across two agent frameworks, showing that basic task completion doesn’t reliably predict metamorphic stability.

I also build explainable systems—most notably an end-to-end deepfake explainability framework that combines visual and natural-language justifications (frame-level Xception features, Grad-CAM, SHAP temporal attribution, and LLM-based explanations). Alongside this research, I’ve applied deep learning to computer vision (e.g., traffic sign detection with real-time webcam + voice output) and translated models into interactive applications (Streamlit), and I’m especially motivated by research that’s rigorous, reproducible, and actionable for AI safety.

Experience

Work history, roles, and key accomplishments

Current

Industrial AI/ML Intern

Current

Togo AI Labs

Feb 2026 - Present (5 months)

Conducted research on frontier AI topics and produced first-author papers including AgentMorph for metamorphic testing of tool-using LLM agents and work on knowledge localization/editability in small language models. Mentored bootcamp students on experimental methodology, paper writing, and submission workflows.

Metamorphic Testing LLM Evaluation Small Language Models Mentoring

Research Intern

GES’s R. H. Sapat College of Engineering

Jan 2025 - Aug 2025 (7 months)

Developed an end-to-end explainability framework for deepfake detection combining visual and natural language explanations to address black-box limitations. Implemented an interactive web application for real-time deepfake detection with detailed explanation and visual interpretability.

Deepfake Detection Explainer Graphics Xception Grad CAM SHAP Transformers

Transformers for Image Detection

GES’s R. H. Sapat College of Engineering

Jul 2024 - Dec 2024 (5 months)

Conducted research on transformer-based model architectures for advanced image processing and detection tasks, focusing on core components and self-attention mechanisms. Investigated deep learning methodologies for image detection and compared their performance against vision transformer approaches.

Vision Transformer Self Attention Computer Vision Deep Learning

Intern

SachiTech

Aug 2022 - Sep 2022 (1 month)

Worked on developing applications using Java and JavaScript-based web development frameworks, contributing to real-world technical projects. Gained experience in software development and quality analysis through applied project work.

Java JavaScript CSS Tailwind Bootstrap Web Development Software Development Quality Analysis