AI that does the work — not just the talking.
Xiaoye Technology builds AI-native digital workers for insurers, banks and brokers in Hong Kong. Grounded in your own documents and core systems. Every answer traceable to the clause it came from. In production in ten weeks.
Three things insurers keep asking for
We are not a general-purpose chatbot vendor. Every system we ship is built on the client's own policy wordings, SOPs and business systems.
Insurance AI assistant
A service assistant for agents and frontline teams that genuinely understands policy wordings, coverage, underwriting rules and claims procedures across the full policy lifecycle.
- Intent routing accuracy ≥92%
- Multi-intent dialogue, context retention ≥90%
- Refuses to answer out-of-scope questions ≥95% of the time
Knowledge & compliance engine
A hybrid retrieval architecture — large models, vector search and reranking — that turns PDF, Word, Excel and PowerPoint archives into a governed, versioned, auditable knowledge base.
- Every sentence traceable to a clause or internal rule
- Version control, approval workflow, instant publish
- Automatic knowledge-gap detection from missed queries
Business-system agents
Agents that reach past the document layer into policy, claims, payment and customer systems through secure read-only APIs — then close the loop by raising and routing tickets.
- Policy, claims and premium lookup in-conversation
- Auto-generated tickets with priority and routing
- Feishu / WeCom integration for notification and approval
Built for an industry that cannot afford a wrong answer
Financial institutions do not fail AI pilots because the model is not clever enough. They fail because nobody can explain where an answer came from, and because the system cannot touch the systems that matter.
Domain-grounded, compliance-first
Product structures, business rules, operational flows and regulatory constraints are modelled explicitly. Grey areas and disputes trigger a deliberate refusal rather than a guess.
No black box
A three-layer hybrid of large models, vector retrieval and reranking. Recall and hallucination are measured, not asserted, and every response carries its source.
Master–Mate multi-agent architecture
Separate agents for policy enquiry, claims guidance, payment lookup and system operations, each independently enabled or disabled, with a central router handling intent.
Deep integration, both directions
Not a document Q&A box. Secure APIs into core systems let the assistant query live business data and drive workflow — tickets, escalations, follow-up scheduling.
You own the system
A visual knowledge-operations console lets business teams upload and approve content themselves. Model-agnostic by design: switch LLM providers without re-architecting.
Scales without drama
Containerised on Docker and Kubernetes with autoscaling. ≥300 concurrent users, ≥100 requests per minute, sub-2-second page loads, ≥99.5% availability.
Designed against Hong Kong requirements, not retrofitted to them
Every deployment is built to satisfy the PDPO and insurance-sector data governance from day one.
Identity and data protection
SSO authentication enforced before any policy or customer data is shown. HKID, policy and card numbers automatically masked. Fine-grained RBAC throughout.
Data stays in Hong Kong
Hosted in the Hong Kong region with no cross-border transfer. TLS 1.2+ in transit, KMS encryption at rest, sensitive fields redacted in logs. Audit trail retained ≥7 years.
Humans stay in the loop
Complaints, disputes and emotionally charged cases are never handled by AI — they hand off to a human agent carrying the full conversation context.
In production with Hong Kong insurers
Replacing a hotline that was drowning in standard enquiries
A Hong Kong life insurer was running an internal phone hotline to support its sales channel. Call analysis showed the bulk of volume was standardised, repeatable enquiries. We delivered an AI assistant covering product terms, policy and claims lookup, premium enquiry and system operations — with a trilingual knowledge base and automated ticket routing.
Client names are withheld under confidentiality obligations. References available on request.
Start small, prove it, then scale
We deliver through an FDE model — forward-deployed engineers who understand both the financial business and the AI stack, working alongside your team rather than emailing specifications back and forth.
Pick your worst bottleneck
One scenario, fast, with a real measurement at the end. The full fee is credited against the project that follows.
End-to-end build
An FDE team delivers end to end with visible progress every two weeks. You receive source code, documentation and training.
SaaS or private deployment
Ready to use, pay as you go, upgrades included. Multi-tenant public cloud, dedicated VPC or fully on-premise.
An AI-native company, with no legacy to defend
Xiaoye Technology Limited is a Hong Kong incorporated technology company building AI systems for the financial sector. We are AI-native by construction: the architecture was designed for large models and concurrent agents from the first line of code, with no compromises inherited from an older product.
Our team pairs financial domain expertise with applied AI engineering. That combination is the reason we can go from kickoff to production in ten weeks — the hard part of these projects is rarely the model, it is knowing which business rules actually matter.
Current R&D focus: Cantonese voice interaction for insurance — speech recognition of mixed Cantonese-English speech, insurance terminology, and natural end-to-end spoken dialogue.
- Legal name
- Xiaoye Technology Limited
- Chinese name
- 小也科技有限公司
- Incorporated
- Hong Kong, 26 March 2026
- CR number
- 80057946
- Registered office
- Unit 12, 9/F, The Cloud,
111 Tung Chau Street,
Tai Kok Tsui, Hong Kong
Tell us where the bottleneck is.
We will tell you honestly whether AI is the right tool for it.