sushant kumar
building intelligent systems that become real products.
AI builder, systems thinker, and product-minded engineer working across agents, retrieval, voice, and applied machine learning. I care about clarity, leverage, and building software that is technically deep, thoughtfully designed, and built to last.
Recent work spans Simpplr AI, Sequoia, Sprouts AI, SquareYards, and Azuro.
- systems thinking
- applied ai
- product architecture
- design-conscious engineering
get in touch if you are building something ambitious with AI.
highlights✨
selected work
A few systems and products from recent roles.
Simpplr AI / August 2025 - Present
Comms AI
Worked on Comms AI at Simpplr, an AI-native workspace for internal communicators spanning planning, drafting, approvals, and multichannel publishing inside a governed enterprise workflow.
Sequoia / January 2025 - August 2025
AI Assist
Built task-oriented AI assist systems for benefits and people operations workflows, connecting enterprise data, retrieval, and automation.
Sprouts AI / June 2024 - January 2025
Autonomous Interview Agent
Built a production interview system that combined ASR, LLM reasoning, and TTS into a live screening workflow.
writing📝
latest writing
Notes on AI systems, research, and the craft of building with them.
April 5, 2026
Rules I live by
A few reminders on building, clarity, taste, leverage, and the kind of work I want to live by.
August 6, 2025
Understanding GPT-OSS architecture
An in-depth look at the architecture of GPT-OSS, an open-source large language model released by OpenAI on 5th August 2025. It is a MoE GPT-2-style Transformer with 36/24 layers, 128/32 experts with top-4 routing, RMSNorm, Grouped Query Attention + RoPE attn, 131 K context via YaRN, 4-bit MXFP4 packs 120 B on 80 GB & 20 B on 16 GB, and SwiGLU activation.
July 26, 2025
Normalization Techniques in Transformer-Based LLMs: LayerNorm, RMSNorm, and Beyond
Deep dive into the evolution of normalization techniques in transformer-based LLMs, from the trusty LayerNorm to newer variants like RMSNorm, and even experimental tweaks.