AI Solutions Forward Deployed Engineer

Mitra Data Semesta (MDS); Jakarta Pusat; On-site; IDR 180,000,000 - 300,000,000; Full time

Job summary

Venus AI, the AI Deployment Division of PT Mitra Data Semesta (MDS), partners with Indonesian enterprises to embed AI in operations. You’ll sit with clients, diagnose problems they’ve misdiagnosed, and design the system with our founder.

This is a designing-and-building role where you’ll work on two to three engagements in parallel, decide when an agent is needed, define the ontology, and build with staged plans and invariant tests before shipping.

Qualifications

• Shipped an LLM system that real users touched, can explain why it broke or behaved unexpectedly.
• Proficient in Python for production systems, with handling of queues, databases, and APIs.
• Excellent product sense: differentiate technically correct outputs from what a sales manager would send.
• Fluent Bahasa Indonesia and precise technical English.
• Willing to tell a client that their requested solution isn’t what they truly need; based in Jakarta or willing to relocate.

Responsibilities

• Run discovery: read client artifacts and come back with a diagnosis that contradicts the brief.
• Decide what shape the solution should take before assuming it’s an agent.
• Design the ontology: entities, rules, and what the system must never decide alone.
• Build through agentic engineering: staged build plans and invariant test suites.
• Write the evals that decide whether we ship.
• Stay with it after launch: monitor real outputs weekly with the client.

Skills

Shipped an LLM system

Production Python

APIs & queues

Product sense

Bahasa Indonesia + English

Job description

Venus AI is the new AI Deployment Division of PT Mitra Data Semesta (MDS), a data center solutions company now focused on AI infrastructure. MDS builds the layer underneath. We build what runs on top of it.

We put AI into the operational core of Indonesian enterprises. The processes that make or lose them money, not the experiments on the side.

The form follows the diagnosis: autonomous agents that qualify buyers and take orders in Bahasa Indonesia, structure pulled out of documents nobody has time to read, workflows automated out of spreadsheets and group chats, internal tools that answer what a company's systems can't. We work across automotive, distribution, healthcare, financial services and retail.

None of it is demoware. Our systems write real orders and hold real customer relationships for weeks. When one gets something wrong, a deal is lost and a director calls us.

The role

You sit in the room while a client describes a problem they've misdiagnosed. You work out what's actually broken, design the system alongside our founder, and build it. Then you watch it meet real users and find out how wrong you both were.

• Run discovery: read the client's real artifacts and come back with a diagnosis that contradicts the brief
• Decide what shape the solution should take before assuming it's an agent, because often it isn't
• Design the ontology before anyone writes a prompt: the entities, the rules, what the system must never decide alone
• Build through agentic engineering: staged build plans, invariant test suites, hooks
• Write the evals that decide whether we ship
• Stay with it after launch: score real outputs weekly with the client and move the numbers

You’ll run two or three engagements at once in unrelated industries. That’s the fastest way to learn what’s genuinely universal versus what every client insists is unique about them.

This is a designing-and-building role. You won’t run a division or carry a number. Pricing, senior client relationships and final go/no-go stay with the founder. Solution design is shared: you’re not receiving specs, you’re in the room where they get written.

Requirements

• You've shipped an LLM system that real users touched, it broke, and you can explain precisely why
• Production Python, and the judgment to reason about queues, databases, retrieval and APIs that lie to you
• Product taste: you know the difference between a technically correct output and one a sales manager would actually send
• Fluent Bahasa Indonesia and precise technical English
• Willing to tell a client the thing they asked for isn't the thing they need
• Based in Jakarta or willing to relocate

Not required: ML degree, research, fine-tuning or MLOps experience, or domain expertise in any industry. You’ll learn a new one every quarter anyway.

You’ll dislike this

if you want a clean backlog, a PM between you and the customer, or six months of research before shipping to a paying client. We don’t train models.