👋 Hi, I'm Shivam Chauhan

🧑‍💻 About Me

I'm an ML Engineer with 5+ years of experience building and deploying production-scale AI systems at Microsoft and MBZUAI. I specialize in multilingual LLMs, generative models, and scalable NLP pipelines. My strength lies in bridging research and real-world applications with a focus on fairness and global representation.

At MBZUAI, I led the development of Nanda — a 10B Hindi-centric LLaMA-3 model — and contributed to multiple research efforts on fairness in generative music models, with first-author publications at NAACL, CACM, AIES, and ACMM. I’ve also designed training pipelines using Azure, SageMaker, and DeepSpeed/FSDP for efficient LLM scaling.

At Microsoft, I spearheaded the integration of multilingual LLMs into Bing Ads, delivering a 20% boost in ad coverage and reducing the bad match ratio by 30% — directly impacting global revenue across EMEA and APAC markets.

🎯 I’m currently exploring roles in Europe where I can bring my end-to-end expertise—from LLM design to deployment—into real-world systems that are inclusive, robust, and production-ready.

📚 Selected Publications

• Llama-3-Nanda-10B-Chat — 10B Hindi LLaMA-3 model (+8% MMLU) [COLM 2025, under review]
• Music for All — On fairness in generative music models [NAACL 2025]
• Missing Melodies — On Global South underrepresentation [CACM, under review]
• Who Gets Heard? — Rethinking fairness in GenAI music [AIES 2025, under review]
• Survey of Music Datasets — Taxonomy & bias in music AI [ISMIR 2025, under review]

💼 Work Experience

🏢 MBZUAI, Abu Dhabi — Senior ML Research Engineer (May 2024 – Present)

• Developing inclusive music generation systems (MusicGen, Mustango) with a focus on global genre diversity
• Lead architect of Nanda LLM (Llama-3-10B), optimized for Indic languages
• Designed full-stack ML pipelines using AdaptDL, SageMaker, and FSDP

🏢 Microsoft, Bengaluru — Data & Applied Scientist II (July 2020 – April 2024)

• Deployed multilingual LLMs in Bing Ads; boosted ad coverage by 20%, reduced bad match rate by 30%
• Led development of ad retrieval and recommender systems; improved CTR and CPC metrics
• Trained lightweight models using Azure + DeepSpeed, cutting training cost by $180K/year

🛠️ Technical Skills

Languages & Frameworks

Python, C++, SQL, Bash • PyTorch, HuggingFace, Keras, Scikit-learn

Platforms

Azure, AWS SageMaker, NVIDIA NGC, Git, Linux

Expertise

LLMs, Adapter-based Fine-tuning, Prompt Engineering, Transfer Learning, Retrieval Models
Evaluation (lm-eval-harness), Multilingual NLP, Low-Resource Modeling, GenAI, Fairness in AI

🏆 Achievements

• 🏅 Microsoft Innovation Award for Bing Ads LLM system (2021)
• 🥈 Microsoft Poster Competition Runner-Up (2022)
• 🎯 AIR 14 – GATE CS (2018)

📢 Connect With Me

Email: 0shivam33@gmail.com
LinkedIn: shivamiisc2018
GitHub: shivam-MBZUAI
Google Scholar: Scholar
Resume: Download CV (PDF)