Artificial Intelligence Engineer

Madison Pearl Ltd.; Hong Kong; On-site; HKD 900,000 - 1,200,000; Full time

Job summary

Madison Pearl Ltd. is seeking an AI engineer to turn customer conversations into rapid POCs and evolve reusable platform capabilities.

You will design multi-agent pipelines for documents, compliance checks, and BQ verification, and build the intelligence layer to embed domain and compliance knowledge into every workflow. You will benchmark frontier models against domain tasks, develop retrieval and reasoning layers, and maintain internal AI agent libraries across client workflows.

Qualifications

• 2+ years hands-on AI/ML engineering & 10+ years in software development.
• Strong in LLMs, RAG, multi-agent systems, and structured output design.
• Python-first development with LangGraph or equivalent agent orchestration.
• Experience building orchestration loops (tools, state, retries, human-in-the-loop).
• Willingness to experiment with new models/frameworks and report findings.
• Startup mindset with ability to operate in ambiguity and ship without a perfect spec.
• Experience with document understanding, OCR pipelines, or structured compliance logic is a plus.
• Comfortable working in Python and a TypeScript web app in the same week.

Responsibilities

• Turn customer conversations into focused, working POCs on a short turnaround and identify reusable platform capabilities.
• Design and ship multi-agent pipelines for document generation, compliance checking, and BQ verification.
• Build the intelligence layer that embeds deep domain knowledge into workflows.
• Benchmark frontier models (GPT, Claude, Gemini) and own the evaluation framework.
• Develop retrieval, reasoning, and structured output layers for auditable documents.
• Create and maintain the internal skill library on the Mink platform for reusable AI agents.
• Integrate CAD/drawing inputs with document output pipelines and cross-reference BQ items and safety plans.
• Lead bi-weekly research reviews to stay at the frontier and implement what matters.
• Collaborate with QS, safety engineers, and civil engineers to encode compliance logic into AI workflows.

Skills

LLMs

RAG

Multi-agent

Python

LangGraph

Agent orchestration

Experimentation

TypeScript

Tools

LangGraph

TypeScript

OCR pipelines

Job description

On behalf of a well-funded and rapidly growing early stage AI development company focused on the compliance space we are seeking an AI engineer who is excited by rapid experimentation and deep platform R&D.

Role:

• Turn customer conversations into focused, working POCs on a short turnaround, then identify what should become a reusable platform capability
• Design and ship multi-agent pipelines for document generation, compliance checking, and BQ (Bill of Quantities) verification
• Build the intelligence layer that brings deep domain and compliance knowledge into every workflow, rather than solving the same problem customer by customer
• Benchmark frontier models (GPT, Claude, Gemini) against domain-specific tasks; own the evaluation framework
• Build retrieval, reasoning, and structured output layers — our documents need to be auditable, not just plausible
• Develop and maintain our internal skill library on the Mink platform — reusable AI agents deployable across client workflows
• Integrate CAD/drawing inputs with document output pipelines (cross-referencing drawings against BQ items, safety plans, method statements)
• Run bi-weekly research reviews — we stay at the frontier and implement what matters
• Work with domain experts (QS, safety engineers, civil engineers) to encode deep compliance logic into AI workflows

Requirements:

• 2+ years hands-on AI/ML engineering & 10+ years in software development.
• Strong in LLMs, RAG, multi-agent systems, and structured output design.
• Python-first, hands-on with LangGraph (or a comparable agent-orchestration framework) — you can build, debug, and extend a real graph.
• Agent instincts — you've built or extended an orchestration loop (tools, state, retries, human-in-the-loop), not just called a chat API.
• You experiment before you theorise — new models, new frameworks, you test them and report what's real.
• Startup instincts — you operate in ambiguity and ship without a perfect spec.
• Document understanding, OCR pipelines, or structured compliance logic — a real plus.
• Comfortable across the stack — happy in a Python and a TypeScript web app in the same week.