Research Engineer (AI Agent Systems)

About Castor

Castor Innovations is an AI-native construction infrastructure company based in Hong Kong and incubated at Cyberport. We build the agent layer for the built environment — turning frontline construction and facilities work into AI-native, auditable, on-site workflows. Our work spans multi-agent systems, on-device vision for site safety, and AI digital workers deployed with real public-sector clients in Hong Kong.

The Role

We are hiring a Research Engineer to advance the core technical frontier of our agent platform. This is a hands-on research and development role focused on resolving genuine scientific and technical uncertainty — taking capabilities from fragile demonstrations to deterministic, production-grade reliability. The work is carried out primarily in Hong Kong.

Key R&D Responsibilities

• Research and resolve the determinism and auditability challenges of multi-agent systems in construction and public-works contexts — including state-machine orchestration, idempotency and de-duplication control, and verifiable “instruction → action → confirmation” task closure.

• Investigate the reliability boundaries of vision-language models (VLMs) for on-site safety detection (PPE compliance, restricted/hazardous-zone entry, work-at-height and machinery operations), solving the technical gap between prototype and production-grade stability.

• Develop accurate automatic speech recognition (ASR) for Cantonese and multilingual voice commands under offline and high-noise construction conditions.

• Design and build the agent runtime: intent routing, state machines, idempotency and de-duplication mechanisms, and provenance for cross-SaaS data exchange.

• Build a domain knowledge base and retrieval-augmented generation (RAG) capability for construction, public-works and regulatory knowledge — researching retrieval accuracy, grounding and provenance to ensure agents answer from verifiable sources rather than hallucinate.

• Establish an agent reliability evaluation (eval) framework that quantifies closed-loop correctness, and iterate toward convergence through controlled experiments and error analysis.

Requirements

• Education: A bachelor’s, master’s or PhD in Computer Science, Artificial Intelligence, Electronic Engineering or a related STEM discipline.

• Right to work: Candidates have full working rights for Hong Kong, or otherwise legally permitted to work in Hong Kong for the full duration of the engagement.

• Core skills: Strong Python and backend/systems engineering; hands-on experience with LLMs, RAG / retrieval systems and/or multi-agent orchestration; and depth in at least one of: state-machine/distributed-systems design, computer vision, or ASR.

• Mindset: Thinks natively in terms of correctness, state and evaluation — not just “getting the model to respond.”

Nice to Have

• Experience hardening LLM agents for production (routing, guardrails, idempotency, regression evals).

• Familiarity with construction / BIM / digital-twin, IoT or edge-device deployment.

• Cantonese language ability (for ASR work and site collaboration).

What We Offer

• A founding-stage research role at the technical core of an AI-native infrastructure company, with real public-sector deployments.

• Direct ownership of hard, defensible problems — the “demo-to-reliable” gap is the moat.

• Monthly salary by qualification: PhD — HK$45,000；Master’s — HK$23,000；Bachelor’s — HK$20,000.

• This position may be supported under the Innovation and Technology Commission’s Research Talent Hub for eligible candidates.

How to Apply

Please send your CV and a short note on a hard agent / reliability problem you have solved to [email protected]. Applications are reviewed on a rolling basis.

Castor Innovations is an equal opportunity employer. All applicants are assessed solely on merit and on the requirements of the role, through an open and fair selection process. Personal data collected will be used for recruitment purposes only.