Shubham Gaur

AI/ML Research Engineer • MS NLP @ UC Santa Cruz

🌍 San Francisco Bay Area, CA, USA
📧 sgaur2@ucsc.edu
📱 (408) 640-5717
🔗 LinkedIn | GitHub | Website | Google Scholar

---

🎯 About Me

Graduate Student in Natural Language Processing at UC Santa Cruz (Silicon Valley Campus) with 5+ years of industry experience at Nokia, Adani Group, and BlackRock. Currently based in Santa Clara, CA, working on advancing generalist AI through multimodal systems and agentic AI.

Research Interests:

• 🎯 Alignment & Reasoning in AI Systems

• 🔧 Efficiency in Large Language Models

• 🤖 Multimodal Systems (text, image, video)

• 🎯 Agentic Systems (Conversational AI, GUI Agents)

---

🎓 Education

University of California, Santa Cruz (UCSC) | Santa Clara, CA

M.S. in Natural Language Processing (Major: AI) | Sept 2024 - Dec 2025

• GPA: 4.0/4.0

SRM Institute of Science & Technology (SRMIST) | Chennai, India

B.S. in Information Technology (Major: ML) | Jul 2015 - May 2019

• GPA: 3.8/4.0

---

💼 Work Experience

🔵 Nokia | Machine Learning Intern | Jun 2025 - Present

Naperville, USA

• Built Graph Neural Network-based anomaly detection models and applied PageRank algorithms
• Developed transformer-based forecasting pipeline for network optimization
• Created agentic multimodal RAG pipeline (LangChain + Qdrant + Azure ML) for enterprise knowledge retrieval

🟢 Adani Group | Lead Machine Learning Engineer | Nov 2021 - Aug 2024

Gurgaon, India

• Built a 100M+ user real-time Customer Data Platform with 10-member team, powering personalized AI/ML applications across travel, retail, and finance
• Developed AdaniOne Chat, an LLM-powered multi-tenant chatbot with Delta Lake-backed vector search
• Deployed GPT-3.5 SFT models for automated hotel description generation across 33K listings (saved $130K)
• Applied transformer-based models (BERT, GPT) for feedback analysis and churn prediction: improved customer retention by 48% and CLV by 15%
• Developed recommendation system with advanced ranking techniques: increased duty-free engagement by 87% and revenue by 20%

🔴 BlackRock | Machine Learning Engineer | Jul 2019 - Oct 2021

Mumbai, India

• Developed neural network-based solution for portfolio and index tree generation, scaling to 100 portfolios
• Built and deployed ETL tools for data validation and securities loading, saving time equivalent to 7 FTEs
• Designed XGBoost and Random Forest-based models for risk scoring and ranking investment instruments

---

🔬 Research Projects

🤖 GUI Agents (in collaboration with Samsung Research America)

May 2025 - Present

• Building web action agents combining DOM, accessibility trees, and reinforcement learning (PPO) with visual cues
• Developing reliable multi-modal UI automation systems
• Creating benchmarks for web-based task generalization

🏠 HomeHelper - Multimodal Agent for Appliance Troubleshooting

Apr - June 2025

• Multi-agent system integrating images, text, and video for household appliance support
• LangChain orchestrator-based system with Self-Improving RAG loop (SIM-RAG)
• Models: BLIP, GPT, Mistral, LLaMA with memory for conversations
• Dataset: 3,000+ synthetic multi-turn conversations

📊 Evaluation of Faithfulness over ARMs and Diffusion Language Models | Code

Apr - June 2025

• Benchmarked diffusion-based LLMs (LLaDA) against ARMs (LLaMA-3, SmolLM) across summarization datasets
• Demonstrated LLaDA’s robustness through zero-shot, chain-of-thought, and adversarial prompting
• Used BERTScore (semantic) + AlignScore (faithfulness) for comprehensive evaluation

🎨 Leveraging LLMs and VLMs for Idiomatic Understanding | Code

SemEval’25 Research

• Used LLMs & VLMs (OpenCLIP, BLIP, ALIGN) to interpret idiomatic sentences
• Developed image-text alignment strategies for contextual meaning interpretation
• Publication: Submitted to SemEval 2025 (ACL Workshop)

---

📝 Publications

1. “Leveraging LLMs and VLMs for idiomatic understanding” | ACL’25 Workshop (SemEval) Paper
   Judith Clymo, Adam Zernik, Shubham Gaur

2. “Advancing Web-Based Visual Question Answering with Efficient Image-Text Alignment” | ICRAAI’24 Paper
   Saketh Kilaru, Shubham Gaur, Spandan Rout

3. “Extraction of Cumulative Blobs Using Dynamic Gestures” | IJSR’21 Paper
   Rishabh Naulakha, Shubham Gaur, Dhairya Lodha, Mehek Tulsyan, Utsav Kotecha

---

🛠️ Technical Skills

Programming & Data Science

Libraries: pandas, NumPy, SciPy, scikit-learn, Matplotlib, Jupyter

Machine Learning & AI

Specializations: Computer Vision (CNN, OpenCV), Reinforcement Learning, LoRA/PEFT, Diffusion Models

Generative AI & LLMs

Models: BERT, GPT-3/4, LLaMA, CLIP, ALIGN
Frameworks: LangChain/LangGraph, AutoGen, Retrieval Augmented Generation
Vector Databases: Pinecone, Qdrant
Techniques: Supervised Fine Tuning, RLHF, PPO, Instruction-Tuned LLMs

Big Data & Cloud

Technologies: Apache Spark, Hadoop, Kafka, Databricks, BigQuery, Synapse, Data Factory

Deployment & MLOps

Tools: FastAPI, MLflow, Airflow, CI/CD, Redis, PostgreSQL, MySQL, MongoDB

---

🏆 Key Achievements

• 🚀 Built 100M+ user platform at Adani Group
• 💰 Saved $130K through automated content generation
• 📈 Improved customer retention by 48% using AI/ML models
• 💼 Increased revenue by 20% through recommendation systems
• ⚡ Saved equivalent of 7 FTEs through automation at BlackRock
• 🎓 Maintained 4.0 GPA in graduate studies
• 📚 3 research publications across AI/ML conferences

---

📞 Let’s Connect!

I’m always interested in discussing AI research, collaboration opportunities, or innovative projects. Feel free to reach out!

• 📧 Email: sgaur2@ucsc.edu
• 💼 LinkedIn: shubhamggaur
• 🐙 GitHub: shubham2345
• 🌐 Website: shubham2345.github.io
• 🎓 Google Scholar: Profile

---

“Advancing AI systems that understand, reason, and act across multiple modalities with human-level reliability.”