I build dependable AI products, from infrastructure to interface.
I'm Sarthak Agrawal, an AI infrastructure and product engineer. I turn model capabilities into systems people can trust: local code review, post-training tools, private assistants, research signals, and the operating layer behind them.
// current focus
Four products, one platform.
Four focused products — CodeVetter, PostTrainLLM, HeyPace, and High Signal — with SaaS Maker as the directory and operating layer for everything else.
CodeVetter
Local-first desktop AI code review for agent-generated code.
view product →PostTrainLLM
A Mac-local LLM factory, runtime, and WebGPU learning playground.
view product →HeyPace
An on-device Mac voice agent with screen-aware local context.
view product →High Signal
An evidence-first daily brief for technology, markets, and infrastructure.
view product →SaaS Maker
The broader directory and operating layer for the full product fleet.
open the directory →// selected work
A few things I've built.
Four projects from my day jobs — what the problem was, what I built, and what changed.
PostTrainLLM — a transformer built end to end
pytorch · webassembly · webgpu · browser
A small GPT (~0.8M params, byte-level) written from scratch in three layers — a PyTorch reference, a hand-derived C++/WASM implementation, and a full WebGPU training loop — running entirely in a browser tab. Every layer's backward pass was finite-difference checked before being trusted.
Vector-Powered Personalized Feeds
embeddings · vector search · ranking
A home feed that actually learns what you like. Content gets embedded with BERT, ranked by similarity in Milvus, and the user vector keeps updating from live events. Engagement went up 40%.
Real-Time Market Data Pipeline
streaming · protobuf · fan-out
The streaming backbone of a fintech social app. Go services push live market data through Kafka to clients in real time — and it held up while daily users went from 15k to 200k.
RAG Agents for Support & Learning
retrieval · openai · moderation
A set of RAG chatbots — support, learning, assistant — built on OpenAI APIs and grounded in real product docs. The support one cut human replies by 90%.
Durable Financial Workflows
temporal · reliability · go
Financial planning can't run on workflows that quietly break. Moving them to Temporal killed 90% of the random failures and gave the team back about three hours a day.