about

last updated january 2026

hey @everyone, swapnil here.

i’m a just a nerd who loves building things that look and feel great and or pique my interest in a particular topic.

this website is a living record of my work, projects, musings, curiosities and research. it will be updated as i progress in my journey. in the very near future, i will also use this site for collecting articles, videos and bookmarks.

currently interested in

• VLM architectures and multimodal reasoning systems
• LLM model architectures, pre-training, post-training, and inference optimization
• Agentic systems (tool-calling, memory, and evaluation)
• RL environments and agent training loops (GRPO/RLVR)
• Cognitive architectures for agents (hybrid memory with knowledge graphs + vector dbs)
• High-performance ML systems and reproducible infra (nix, pixi)

currently working on

• openenv contributions focused on multi-session reliability and scalable rollouts
• an openenv-compliant computer control environment for desktop task learning, using trl with grpo
• training a small language model from scratch
• this website

not work

• In my free time, i hop on valorant and marvel rivals with my friends
• i also like to indulge in some digital art from time to time

contact

• you can reach me at [email protected]
• fairly active on discord @swappy._.