AI Engineer - 3echo Workspace

3ECHO PTE. LTD.; Singapore; On-site; SGD 90,000 - 130,000; Full time

Job summary

3ECHO PTE. LTD. in Singapore seeks an AI Engineer to ship end-to-end AI-powered workflows, agents, and client Workspaces. You’ll work directly with the founder in a small, fast team delivering real business outcomes for paying clients.

The role emphasizes shipping real systems with Claude-centric tooling, MCP and integration layers, and a strong emphasis on practical, production-ready code. Expect 2–4 years of hands-on experience in a Claude-native environment.

Qualifications

• Mid-level: 2–4 years shipping real systems.
• Experience shipping end-to-end AI/LLM-powered products is preferred.
• Familiar with Claude ecosystem and tool-calling APIs.
• Proven backend production ownership with reliability and observability.

Responsibilities

• Own agent/workspace infrastructure and workflow orchestration.
• Develop and maintain the generative media pipeline and related tooling.
• Deliver full-stack features for client Workspaces using Python/FastAPI and TypeScript/React.

Skills

Python / FastAPI

TypeScript

React

Tools

Python/FastAPI

TypeScript

React

Postgres/Supabase

Docker

GCP Cloud Run

Job description

Company Overview

3echo is a Singapore-based AI company building agent-powered products for businesses across Southeast Asia. We deploy AI agents that do real work - content production, operations, client delivery - measured by business outcomes, not features shipped. Small team, founder-led, fast iteration, paying clients.

The Role

You're an AI Engineer who ships end to end. Not a prompt tinkerer, not a pure backend dev - you sit where the agent workflow, the generative media pipeline, and the client's actual business outcome meet. You'll work directly with the founder in a small, fast team where your code goes to paying clients in days, not quarters.

Mid-level: 2-4 years shipping real systems. What matters more than years is that you've built things people use and that Claude is how you work. We're a Claude-native shop: Claude Code is the primary development environment, MCP is the integration layer, and agent skills are how we encode process. One engineer here ships what a team of five ships elsewhere, because the leverage is the tooling. If you're not already working this way, you should be hungry to.

What You'll Own

1. Agent & Workspace infrastructure

   • Build and maintain MCP servers, agent skills, and workflow orchestration that power client Workspaces
   • Encode business rules - budget gates, approval SLAs, credit accounting, sign-off rituals as self-enforcing agent behavior
   • Automation pipelines in n8n connecting our register, comms, and client channels

2. Generative media pipeline

   • Operate and extend the Studio engine: image generation (gpt-image-2), video generation (Seedance 2.0 via BytePlus), reference-anchored character consistency, take/cut/release pipelines
   • Cost-aware generation: quota management, credit metering, resolution strategy, provider failover
   • QA tooling for identity hold, canon compliance, and output quality

3. Full-stack product & client delivery

   • Backend services in Python/FastAPI and TypeScript, Postgres/Supabase, deployed on GCP Cloud Run
   • Client-facing surfaces in React where the Workspace needs a UI
   • Forward-deployed work: onboarding client Workspaces, debugging live jobs, turning client feedback into shipped fixes fast

Our stack

Claude & Claude Code - MCP - n8n - gpt-image-2 - Seedance 2.0 (BytePlus) - Gemini (OpenRouter) - Python/FastAPI - TypeScript/React - Postgres/Supabase - Docker - GCP Cloud Run - on-prem RTX 5090 GPU servers

You don't need to know all of it. You need to be dangerous in most of it within your first month.

You've probably done some of this

• Shipped an LLM-powered product or agent system to real users (side projects count if they're live)
• Built with tool-calling / MCP / function-calling APIs, not just chat completions
• Worked with generative image or video APIs and dealt with their failure modes
• Owned a backend service in production - deploys, logs, incidents, the unglamorous parts
• Ship daily with Claude Code (or equivalent) as your primary environment - you know how to scope tasks for an agent, review its output critically, and get 5-10× throughput from it, not 1.2×
• Written system prompts, skills, or MCP tools that made a model reliably do a job - you treat the model as a system to engineer, not a chatbot to talk to

How we hire

We hire on AI DNA : attitude, learning curve, hunger. Certifications and brand-name employers don't move us. A GitHub repo, a live demo, or a story about the hardest thing you shipped moves us a lot. Attitude disqualifies faster than ability qualifies.

Part of the interview is watching you work with Claude on a real task. We're not testing whether you can code without AI, we're testing how much more you can do with it.

First 90 days - what success looks like

• Week 2: shipped your first fix to a live client Workspace
• Month 1: own one of the three pillars day-to-day with the founder reviewing, not driving
• Month 3: a client outcome (turnaround, first-pass rate, or cost-per-deliverable) measurably improved by something you built