Builder profile

Raghav Agarwal

@2008raghavagarwal

CS at UIUC researching LLM reliability, covering error cascades in multi step reasoning, memorization in diffusion training, and multi agent value decomposition. Five first author ICML 2026 workshop papers, plus IEEE PowerAfrica and RANLP. Building a testbed that scores monitors on rogue agents.

• Python
• C++
• PyTorch
• Hugging Face Transformers
• CUDA
• vLLM
• LoRA fine tuning
• AWQ quantization
• distributed training
• SLURM and HPC
• LLM evaluation
• benchmark construction
• agent security
• prompt injection
• mechanistic interpretability
• residual stream probes
• activation steering
• RAG
• Docker