Baljinder S. Hothi
Some things about me:
- MTS FDE Infra at Cohere, working on deployment, GPUs, scaling, and all things infrastructure
- Previous SDE intern at AWS, member of Serverless Lambda, worked on an open-source MCP server
- Previous Engineering Fellow at Meta, investigated humanoid agent manipulation failures in HITL sandbox
- Undergraduate Researcher at RAIVN Lab (University of Washington), benchmarking behavioral cloning methods for multi-task policy learning under Nabil Omi
- ML Intern at Interdisciplinary Data Science Lab, built heart disease classifier with LMCLUS achieving 87% accuracy
- Software Engineering Intern at ICCAE, developed LLM + TTS for robotic navigation and trained NLP model to detect LLM-generated text with 80% accuracy
Some things I wrote:
Some projects I worked on:
- Tiny-Llama-Optimization —inference acceleration framework for TinyLlama-1.1B using mixed precision, INT8 quantization, attention head pruning, and KV-cache quantization
- hyrax-lib — lightweight distributed training across multiple datasets on local hardware, no Kubernetes or cloud required
- cooperative-push-marl — two MuJoCo robots learning to coordinate and push objects too heavy to move alone
- Gungir — Async distributed RL pipeline supporting continuous control (MuJoCo) and LLM post-training under a unified stateless-worker architecture.
Some things I shot:
Some places to find me:
- Twitter: @baljhothi
- Github: @BaljinderHothi
- LinkedIn: @baljinder-hothi
Edited March 03, 2026