Full Stack Engineer

About Manufex
Manufex is the system of record for manufacturing procurement. Our tagline is "every order, on the record" and we mean it literally: buyers and suppliers run real commercial work through the platform, so there's an actual record of what was agreed and when.
We're a small team. You'd ship to production in your first fortnight.
About the role
You'll build real features end to end, from the API through to the interface. You won't be handed tickets to close. You'll get a problem, help decide what it should do, and build it.
What you'll do
Data modelling: design schemas and write migrations that stay safe once real data is live
APIs: design endpoints and work out the states a record can legally move between
Interfaces people use daily: build screens where the hard part is deciding what to leave out
Access control: implement multi-tenant permissions, where a mistake means one customer sees another's data
AI-backed features: apply retrieval and language models to customer content, including what happens when the model is wrong, slow, or down
Integrations and file pipelines: work with third-party services and uploads, and handle the failures that come with both
Our stack
You don't need to know these already. This is what you'd learn.
Backend: Python, FastAPI, SQLAlchemy (async), Pydantic, pytest. Postgres and auth on Supabase. Retrieval on LlamaIndex with Claude and Gemini.
Frontend: React, TypeScript, Vite, TanStack Query, React Router, Tailwind, Vitest.
Everywhere: GitHub, PR review, CI on every pull request, continuous deployment.
How we work
We write things down first. Design docs are the source of truth, and code conforms to them rather than the reverse. Every decision gets recorded with its reasoning: which library, which trade-off, which alternative we rejected and why.
We span two time zones. Part of the team is in San Francisco, twelve and a half hours behind Bangalore. Overlap is real but limited, typically [your evening, their morning], and the rest of the day is yours. It's also why we write so much down. When the person who can answer your question is asleep, the answer has to already exist.
We use AI coding agents daily and expect you to. What matters isn't how fast you generate code. It's whether you read it, understand what you're shipping, and can say why it's correct.
We care about reliability. Passing tests aren't working software. We'd rather you spent twenty minutes proving something works than twenty minutes making it look finished.
Priorities move. We're early. What matters this month may not matter next month.
What we're looking for
You've built something that works, front and back. A personal project, coursework, an internship, anything you can walk us through
You build AI-natively. Agents are part of how you work. You know what to delegate, what to check yourself, and what it means when a model sounds certain and is wrong
You're curious about your tooling: skills, MCP servers, hooks, custom commands. You find a better way to work and tell the rest of us about it
You deliver when priorities shift. Plans change here, often. You re-aim without stalling, and things still ship
You're comfortable in TypeScript or Python. One is enough
You ask questions when a request is ambiguous instead of guessing
You can explain your own code, including the parts an AI wrote
You can prompt clearly and own documentation. Much of this job is explaining decisions to other people
You can work in person in Bangalore, or you're ready to relocate here

Work arrangement
No