NOBODY NAMES THE TASK
The team cannot say which piece of work should change first, or who owns the change. Ambition stays abstract, so nothing ships.
AI ADOPTION that generates REAL ROI
We help Hong Kong teams put AI into one workflow they already run, then build agents that take the routine work off it, on the systems they already use. With a private, on‑premise path when the data demands it.
BOOK A WORKING SESSIONWORKING WITH TEAMS ACROSS HONG KONG
PILOTS THAT STALL
Proofs of concept impress in the demo, then never reach the daily work
YOU BOUGHT THE TOOLS. adoption didn’t follow.
Most teams do not need another license or another training day. They need one task done a better way, clear rules for data, and proof that the new way beats the old one. Four things usually stand in the way.
NOBODY TIMED THE OLD WAY
No one wrote down how long the work took before. So no one can show what got better, and the budget dies at the next review.
THE DATA RULES ARE UNCLEAR
Nobody says what may go into which tool. People paste too much into public models, or avoid AI completely. Both cost you.
THE PILOT NEVER LANDS
A demo works in a sandbox and stalls before production. The gap between a pilot and daily use is where most AI budgets go to die.
That is why we start small: one task, the team that owns it, and a number to beat.
ONE WORKFLOW, across every system.
A quote starts in the CRM, becomes an invoice in accounting, then a folder in file storage and an email to the client. We build agents on top of the systems you already pay for, so the workflow runs as one. We will never make you migrate off your systems. If you want to, we will help you through it.
BUILT ON WHAT YOU OWN
Your current systems stay the system of record. Agents read and write through their official interfaces. Nothing is migrated, nothing replaced.
ONE FLOW, END TO END
The agent carries a task across your tools: the same record moves from quote to invoice to filing, without being retyped along the way.
A PERSON STAYS IN CHARGE
Agents draft, reconcile, and prepare. A named person approves what matters. Every action is logged, every rule written down.
ONE TASK at a time.
Every engagement runs the same way, from a two‑week first deployment to a private setup. Plain steps, a fixed fee, and a clear answer at the end: expand, revise, or stop.
DAYS 1–2
PICK THE TASK
Choose one task where a better way matters. Agree who uses AI, what data is off limits, and the number to beat.
DAYS 3–8
RUN IT WITH THE TEAM
Set up the tools and the rules, then work on live cases together until the new way holds without us in the room.
DAYS 9–10
CHECK AND DECIDE
Put the new numbers next to the old ones and agree what happens next: expand, revise, or stop. If the case is weak, we will recommend stopping.
You keep the working setup, the rules, and the training, whatever you decide. THE BASELINE DECIDES, NOT THE DEMO.
FIXED FEES, agreed in writing.
Every engagement runs on a fixed fee, agreed in writing before work begins: no hourly billing, no open‑ended retainers. Software, cloud services, and hardware are billed separately unless the proposal says otherwise.
MOST TEAMS START HERE
First Deployment
Two weeks
ONE TEAM, ONE TASK · FIXED FEE
- The old way timed before anything changes
- Tools and data rules set up with the team
- Live cases run together until the new way holds
- A clear answer: expand, revise, or stop
AFTER PROOF
Embedded Adoption
8–12 weeks
ONE TEAM · FIXED FEE
- We stay until the new way is routine
- Continuous improvement on live work
- Training for the people who do the task
- Full handover, documented
PRIVATE PATH
Private AI
4–5 week setup
FIXED SETUP FEE · OPTIONAL GATEWAY
- Approved accounts and written data rules
- Optional gateway: masking, logs, spend caps
- On‑premise model, quoted for your site
- You hold the keys; our access is revocable
FOR WORK THAT cannot leave THE BUILDING
Most teams only need approved accounts and clear rules. When the data demands more, we add a gateway you control, or a model on hardware you own. Use the level of control the work actually requires.
FOR TEAMS THAT NEED APPROVED ACCESS FAST
Approved tools
Accounts in your name, written rules for safe use, staff training, and a plan for daily work.
FOR TEAMS THAT NEED MASKING AND LOGS
A gateway you control
Your team reaches external models through one gate in your cloud account. It masks sensitive data, logs use, and caps spend.
FOR WORK THAT MUST STAY ON‑SITE
A model on your hardware
A private model on machines you own, tested on your real workload, with a written runbook for your team.
You keep the accounts, the keys, the logs, and the hardware. You can remove our access at any time.
AN INDEPENDENT practice in Hong Kong.
AlpinTech was founded by Johannes Janousek. Engagements are kept small on purpose: one team, one task, and a clear answer at the end.
He brings years of experience as an AI engineer and data scientist, including at Massar Capital, the New York macro hedge fund that won the HFM award for best macro performance. He holds an MSc in Data Science from King’s College London, held startup roles in Berlin and Hong Kong, and is an Anthropic‑certified Claude Code Architect.
The practice is independent: no reseller agreements, no commissions, and no software of its own to sell. Recommendations are limited to what fits your workflow and your data policy, and every engagement is documented so the work remains useful after it ends.
- MASSAR CAPITAL, NEW YORK
- Data scientist at the HFM‑award‑winning macro fund
- KING’S COLLEGE LONDON
- MSc Data Science
- BERLIN · HONG KONG
- AI engineering and startup roles
- ANTHROPIC
- Certified Claude Code Architect ↗
ASKED FIRST, answered plainly.
Who is this for?
Teams whose work is analysis, documents, operations, code, or client service. Not only engineers, and not only large companies.
How do we know it works before we spend more?
Start with the First Deployment. It is a small test on your own work, checked against your old numbers. If the results are weak, we will say stop. The first deployment is the decision, not the commitment.
How do you handle data and security?
We agree the rules in writing before we start: which tools are approved, what may go into a model, and who reviews the output. If your policy asks for more, the private AI options add masking, logs, and on‑site models.
What happens after the first deployment?
If the numbers hold, most teams expand to a second task or move to Embedded Adoption, where we stay until the new way is routine and then hand everything over. Launching is not the finish line; sticking is.
Do you work in person or remotely?
Both. We are based in Hong Kong and work on site with local teams. For teams elsewhere, we do the same work remotely.
TELL US ABOUT the team and the task.
A short note is enough: what the team does and what you want to change. Kept confidential; no sensitive data is needed at this stage.