Skip to content
Open to AI infra & product roles

I build AI infrastructure — and the products that run on it.

I'm Sarthak Agrawal. Backend-heavy, full-stack — I build LLM tooling, real-time pipelines, and the infrastructure that keeps AI features dependable in production. I also ship the products that run on top.

View selected work About me
sarthak@infra — zsh — 80×24
guest@sarthak:~$ whoami
sarthak agrawal — ai infrastructure & product engineer
guest@sarthak:~$ cat focus.txt
llm tooling · real-time pipelines · vector search · shipped products
type 'help' for commands — or press ⌘K to navigate
guest@sarthak:~$try 'help'
4+
years shipping
200k
daily actives served
101
public repositories
6
languages shipped

// how i think

It's the stuff around the happy path.

Most of the job isn't the code that works first try. It's the request that times out, the queue that backs up, the dependency that goes down. That's what I build for.

~/architecture — request lifecycle
Clients web · mobile Edge / CDN cloudflare API Gateway auth · routing Kafka event stream Vector Search milvus · embeddings MySQL primary store Workers go services Redis cache · realtime
ingress compute stateful

// selected work

A few things I've built.

Four projects from my day jobs — what the problem was, what I built, and what changed.

01 2026

TinyGPT — 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.

~0.8M
params, trained in-browser
24/24
GPU kernels parity-checked
1.6×
WASM SIMD speedup
PyTorch C++ WebAssembly WebGPU / WGSL TypeScript read case study →
02 2024

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%.

+40%
home-feed engagement
Real-time
user vectors
Milvus
ANN vector search
Go Milvus BERT OpenAI / GPT BigQuery read case study →
03 2023

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.

15k → 200k
DAU in 14 weeks
600 → 60ms
page build + load
92%
fewer session DB calls
Go Kafka Protocol Buffers Socket.io Redis read case study →
04 2024

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%.

-90%
human support load
3
agent surfaces
Grounded
retrieval-backed answers
Node.js OpenAI APIs RAG Vector retrieval read case study →
05 2025

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.

-90%
unexpected failures
~3 hrs/day
engineering time recovered
Durable
execution guarantees
Go Temporal MySQL read case study →

// open source

Selected open-source work.

A few favourites, pulled out of 100-plus public repos.

all projects →

reel maker

1

Hand it a topic and it writes the script, makes the images, records a voiceover and renders the video.

TypeScriptupdated May 2026

stitch

Stitches a few video feeds into one panorama using SIFT feature matching and homography.

Pythonupdated Nov 2021

click vs elastic

I loaded 10M rows into ClickHouse and Elasticsearch three ways each, just to see what’s actually faster.

Goupdated Feb 2026

agentMode

Caches a subreddit and gives you AI summaries of it, with links back to the posts they came from.

TypeScriptupdated Apr 2026

status pulse backend

A status-page API built for multiple tenants — roles, auth, and live updates over websockets.

TypeScriptupdated Jan 2025

stumble backend

The backend for a dating app: matching, chat, notifications and cron jobs, on Redis and Prisma.

TypeScriptupdated Aug 2024

// expertise

What I work on.

From the API down to storage, plus the AI parts that sit on top.

01

Backend & APIs

Getting the service right comes before getting it fast. I care about clean contracts and knowing how things break before they do.

  • Go
  • Node.js / TypeScript
  • Python
  • Protocol Buffers
  • REST APIs
  • Microservices
02

Distributed Systems & Reliability

Streaming pipelines, durable workflows, real-time delivery — the parts that decide whether a product survives getting popular.

  • Apache Kafka
  • Temporal
  • Socket.io
  • Queue-based batching
  • Docker & Kubernetes
  • Prometheus
03

AI / LLM Infrastructure

Personalized feeds, retrieval pipelines, agents. The goal is AI that holds up in production, not just in a demo.

  • RAG pipelines
  • OpenAI APIs
  • BERT embeddings
  • Milvus vector DB
  • GPT-powered features
  • Real-time vector enrichment
04

Data & Storage

Pick storage for how the data actually gets read and written. Then cache the slow paths until they stop being slow.

  • MySQL
  • PostgreSQL
  • Redis
  • ClickHouse
  • Elasticsearch
  • BigQuery

// trajectory

Where I've been.

A short version. The longer story lives on the about page.

  1. Feb 2025 — Present

    Software Engineer · VaultWealth Peak XV-backed

    Backend services and reliability infrastructure for a wealth-management platform — financial planning, durable workflows, and the systems they run on.

  2. Jan 2022 — Jan 2025

    Software Engineer · Front.Page YC S’21

    Backend and data infrastructure for a fast-growing fintech social product — real-time market data, personalized feeds, and the AI systems layered on top.

full background →