AI Builder

Mekari; Jakarta Pusat; On-site; IDR 360,000,000 - 600,000,000; Full time

Job summary

Mekari is building a shared inference platform that connects products to AI via a governed path to external models and hosted systems. This full-stack role owns the API/backend, web control surfaces, models, and drivers, ensuring production readiness under real load.

This role covers platform shipping, intelligence supply, and evaluation, shaping secure REST APIs, streaming endpoints, and policy-aware routing while collaborating with cross-functional teams.

Qualifications

• Experience shipping full-stack platforms with AI components.
• Proven ability to design secure, scalable APIs and services.
• Strong background in backend systems under load and concurrency.

Responsibilities

• Stay close to tasks, acceptance criteria, quality risks, traffic patterns, customer feedback, and incidents.
• Own the request path: API contracts, authentication (OAuth2), RBAC, tenant isolation, policy, routing execution, fallback, streaming, and metering.
• Design and build secure REST APIs and streaming endpoints, and backend services that stay correct under concurrency, partial failure, and traffic spikes.
• Build the web control surfaces that let operators observe state without reading source code.
• Integrate external model providers and self-hosted inference endpoints, maintaining a versioned catalog of models, drivers, capabilities, limits, and prices.
• Measure serving performance and plan capacity to optimize cost.
• Develop small task-specific models, prompts, retrieval flows, and agent drivers exposed via the platform.
• Define task-level quality and unacceptable behavior with product teams and domain experts.
• Mentor other engineers and raise design/code quality through review.

Skills

API design

REST APIs

Concurrency

Security

Observability

Mentoring

Tools

PostgreSQL

Kubernetes

Docker

OAuth2

Job description

We are building the shared inference platform that every product in the company reaches AI through: one governed path to external model providers and to the models we host ourselves, with smart routing on top of it. Through those products it reaches thousands of businesses and millions of users.

The scope is deliberately hybrid. The platform side governs predictable behaviour: APIs, authentication, tenant policy, quotas, metering, routing execution, capacity, reliability, security, observability, and deployment. The AI side governs variable behaviour: task understanding, model selection, prompts, retrieval, tool use, small specialised models, output quality, guardrails, evaluation, and feedback. Routing is where the two meet. A request has to be allowed under its tenant policy, and the option chosen has to meet the task's quality, latency, and cost requirements. A response that succeeds technically but fails the user's task does not count.

This is a full stack role with an LLM engineering core. You will own the API and the backend services behind it, the web control surfaces people operate the platform through, the models and drivers it serves, and how all of it behaves in production under real load.

How The Role Ships End To End

• Platform shipping. Reliable, secure, observable inference APIs, routing policies, provider and self-hosted model connections, tenant controls, metering, and the control surfaces teams use to see and manage them.
• Intelligence supply. Task-specific small models, model configurations, and agent drivers that teams can invoke through the platform. An agent driver combines a model with instructions, retrieval, tools, and validation behind a versioned interface.
• Evaluation and feedback. Task datasets, benchmarks, production monitoring, and release gates, using observed failures and economics to improve models, drivers, and routing decisions.

Core Reponsibilities

• Stay close to the tasks, acceptance criteria, quality risks, traffic patterns, customer feedback, and incidents of the products that use the platform
• Own the request path: API contracts, authentication (OAuth2), role-based access control, tenant isolation, policy, routing execution, fallback, streaming, and metering
• Design and build secure REST APIs and streaming endpoints, and backend services that stay correct under concurrency, partial failure, and traffic spikes
• Build the web control surfaces that make a complex system legible, so an operator can see state and act on it without reading source code
• Integrate and operate external model providers and self-hosted inference endpoints, and maintain a versioned catalogue of approved models, drivers, capabilities, limits, and prices
• Measure serving performance under realistic prompt lengths, output lengths, concurrency, and traffic mixes, then plan capacity and optimise configuration and cost
• Build small task-specific models, prompts, retrieval flows, and agent drivers for well-defined product or customer tasks, then expose them through the platform
• Define task-level quality and unacceptable behaviour with product teams and domain experts, and assemble representative evaluation and holdout datasets
• Compare candidate models and drivers against the current baseline on task success, safety, latency, and cost before shifting traffic
• Implement policy-aware routing, confidence or uncertainty escalation, safe fallbacks, and customer-specific controls where the evidence justifies them
• Use shadow traffic, staged releases, regression evaluations, monitoring, and rollback to introduce new serving or intelligence behaviour safely
• Track production failures and feedback, then improve the whole system: data, prompts, models, tools, policies, capacity, and operational controls
• Model data carefully in PostgreSQL, and measure usage and cost accurately enough to report and bill from
• Protect tenant data throughout the request path, respecting privacy, residency, and model-use boundaries
• Maintain dashboards, alerts, runbooks, incident response, and clear ownership for the platform and the capabilities it serves
• Work with product teams, AI engineers, domain experts, infrastructure, and security to make capabilities reusable across products
• Keep team knowledge written down: technical requirements, API contracts, design notes explaining model choice and routing policy, deployment notes, and post-mortems
• Mentor other engineers and raise the bar on design and code quality through review