Hi, I'm Saumya Chaturvedi!

/ˈsaʊ.mjə/

AI Safety Researcher, learning and iterating through life.

RIGHT NOW

Currently participating in Bluedot Impact Technical AI Safety Project.
Organizing and leading a study group focused on ARENA curriculum
Reading about Persona Vectors, AI Safety for Multi-agent systems and building Autograd from scratch

I am interested in AI Safety Research, to work on impactful AI Safety research that reduces the harm potential of AI systems while amplifying their benefits on all aspects of society. I've been a Research Scholar at Max Planck Institute for Software Systems. My work on code-aware language models has been published at ICLR 2025 and NeurIPS 2025, and I presented both papers in person as posters. Currently a Software Engineer at American Express.

Selected research

NeurIPS 2025 · DL4C workshop

SQL-of-Thought: Multi-agentic Text-to-SQL with Guided Error Correction

. Presented as a poster, in person.

arXiv 9 citations poster

ICLR 2025 · DL4C workshop

LoRACode: LoRA Adapters for Code Embeddings

. Presented as a poster, in person.

arXiv 1 citations poster

+ in-progress projects on the research page

From the blog

Bluedot Project Sprint Worklog: Persona Vectors AI safety Transformers from Scratch Research I let Gemini be my Personal Assistant for a month Productivity