hksay

AI agents, grounded in your data · Hong Kong

AI agents over your data.
In production.
Without the 3am page.

We design, ship and run AI agents grounded in your proprietary data and public sources — with retrieval pipelines and eval harnesses that hold up under your workload. CN · HK · TW.

One free audit per quarter, per company. If we can't help, you keep the report.

Get a free infra audit How we work

HKT

on-call coverage

3

languages · EN / ZH / Cantonese

Audit → Embed → Run

one team, all the way through

Representative engagement

One we can describe.
Anonymised at client request.

Most of our work is under NDA. The shape below is real; the client isn't named until they say otherwise. The architecture, the regulator, and the engagement timeline are accurate.

Financial services · Hong Kong · Q1 2025 – ongoing

Compliance Q&A copilot for a HK fintech compliance team

Their compliance team was fielding ~200 internal queries per week from front-office staff: regulatory clarifications, transaction-rule lookups, and KYC exceptions. The team was 8 people, growing headcount got expensive, and the institutional knowledge lived in 14 years of internal memos scattered across SharePoint, email, and three different wikis.

auditembedrun

Read the case study

Headline metric

1,200

queries in the eval harness

anonymised at client request

“We stopped hiring our third junior compliance officer. The copilot handles the body of the work; the team handles the judgement.”

One engagement we can describe, anonymised at client request.

All case studies

How it works

One practice.
Three shapes.
Same team, all the way through.

Most engagements start with the audit. We come back to you in a week with a 4-page report and a 90-day plan. If we can't help, the audit is on us.

01 · 1 week

Audit

We map your data sources (proprietary + public), the retrieval pipeline, the eval harness, and the on-call story. You get a 4-page audit and a 90-day plan.

See audit sample

02 · 4–8 weeks

Embed

Two senior engineers join your team. We ship the missing piece — the retrieval layer, the eval harness, the data integration. To production, on your infra, with your data. Knowledge transfer is part of the contract.

Scope an embed

03 · monthly

Run

We take the on-call. Model swaps, data drift, retrieval quality, monthly architecture review. A real SLA — published monthly, with a runbook behind it. You focus on the product.

Talk about managed ops

Three shapes · one team

Move between tiers as the engagement matures.

See the full comparison

Audit

Free1 week

4-page audit of your data sources, retrieval, evals, and on-call story.

  • Written 4-page audit
  • You keep the report

Request

Embed

Quote4–8 weeks

Two senior AI engineers ship the data integration, retrieval layer, and eval harness with your team.

  • 2 senior engineers
  • Production deploy
  • Source code transfer

Scope it

Run

Monthlymonthly retainer

Managed AI ops with a published monthly SLA. We monitor data drift, retrieval quality, prompt regression, and on-call paging.

  • HKT on-call
  • Quarterly model-swap review
  • Published SLA

Talk to us

One free audit per quarter, per company. If we can't help, you keep the report. The audit has to earn the next step.