Backend Software Engineer

Vocaid; Brasil; Remoto; BRL 180.000 - 290.000; Tempo integral

Resumo da vaga

Vocaid is an AI voice interview platform and a remote-first startup. This role focuses on backend services on the live path of a spoken interview, from speech to text to dialogue decisions, all within tight latency budgets.

You will work with Python 3.12, FastAPI, PostgreSQL, async patterns, and a Rust gateway for the live audio path, and you will own design, tests, deployment, and metrics while shipping to users within the week.

Qualificações

• Production Python web services experience with async frameworks.
• Explain what blocks an event loop and how to debug it.
• Model relational data and evolve schemas with migrations.
• Own a service end to end: design, tests, deploy, metrics.
• Communicate effectively in a distributed async team.

Responsabilidades

• Own backend services on the live interview path, ensuring latency budgets.
• Touch stack: Python, FastAPI, PostgreSQL, and gateway services.
• Collaborate with founder and ship features within a week.
• Maintain observability, tests, and deployments.

Conhecimentos

Python

Async programming

Relational data modeling

End-to-end ownership

Debug performance

Ferramentas

FastAPI

Django

Flask

SQLAlchemy

PostgreSQL

Descrição da vaga

Vocaid is an AI voice interview platform. Candidates hold a real spoken interview with an AI interviewer that responds in the moment, scores the conversation, and returns structured feedback — in any of 10 languages. The same engine powers our B2B hiring product.

We are early-stage and remote-first, and we run the company on our own product: the role you are reading about was posted, and will be screened, through Vocaid itself.

THE ROLE

You will own backend services on the live path of a spoken interview: a candidate talks, speech becomes text, a dialogue model decides what to ask next, and a reply is spoken back. Every part of that loop is on a latency budget a human can feel, which makes ordinary backend decisions — an extra await, a chatty query, a synchronous third-party call — immediately audible.

What you would touch:

• Python 3.12, FastAPI, Pydantic v2
• SQLAlchemy 2.0 (async) and Alembic against PostgreSQL on Azure
• A Rust/Axum gateway carrying the live WebSocket audio path

Problems genuinely open right now, not hypotheticals:

• Holding conversational latency across the ASR → dialogue → TTS chain when the dominant cost is model output tokens, not network round trips.
• Scaling WebSocket interview sessions past a single replica. Session state is not externalised yet, so horizontal scale is real design work.
• Operating inside a deliberately tight PostgreSQL connection budget, where a careless pool setting during a deploy is felt in production.
• Routing model calls through a purpose gate that must respect per-request data-residency rules, so some traffic can never leave a given provider.
• Keeping a scoring pipeline explainable enough to defend to a candidate who asks why they were scored the way they were.

You would work directly with the founder, choose your own tools within reason, and see what you ship reach users the same week.

JOB REQUIREMENTS

• You have built and operated production Python web services — FastAPI, Django, Flask or similar. We care that it ran, was on call, and broke.
• You are fluent in async/await and can explain what actually blocks an event loop, having debugged it rather than read about it.
• You can model relational data and evolve a schema with migrations without taking a write path down.
• You can own a service end to end: design, tests, deploy, metrics, and the page at 2am.
• You write for people who are not in the room — commits, comments, and design notes an async remote team can act on without a meeting.

We do not require a degree, and we do not filter on where you worked. Show us something you built and can reason about — a repo, a write-up, a post-mortem — and that counts more than a title.

DESIRED

None of these are required. Any one of them is interesting to us.

• Rust, particularly Axum or Tokio, or a real appetite to learn it in anger.
• Real-time systems: WebSockets, streaming audio, jitter and backpressure.
• Having integrated ASR or TTS and dealt with the parts vendors do not document.
• Azure Container Apps, or comparable container platforms elsewhere.
• Having worked under a privacy or AI regime — GDPR, LGPD, the EU AI Act — and shipped anyway.
• Experience in an early-stage team where you set the conventions.

HOW WE HIRE, AND YOUR DATA

We use our own product in this hiring process. If you apply, your CV is parsed and scored automatically to produce a match score, and you may be invited to an AI-conducted interview that is recorded, transcribed and scored. That is an automated employment decision tool, and we tell you so up front rather than in the small print.

A human reviews every decision that affects your application. Automated output informs that review; it does not replace it. You may ask for a human-only review, or ask what an automated score was based on, by writing to [e-mail] — asking will not count against you.

How the models are used, and what we do and do not claim for them, is set out at https://vocaid.ai/ai-transparency. What we collect, how long we keep it and how to have it deleted is at https://vocaid.ai/privacy-policy. Where local law gives you further rights over automated decisions or your data, those rights apply and we will honour them.

Vocaid hires on ability. We do not discriminate on race, colour, ancestry, national origin, religion, sex, gender identity or expression, sexual orientation, age, disability, medical condition, marital or veteran status, or any other characteristic protected by applicable law. Tell us at [e-mail] if you need an adjustment to any stage of this process, and we will arrange it.

This posting is not an offer of employment and does not form a contract. Engagement type, jurisdiction and compensation are settled in writing before any offer.