Bring us a hard problem
Who we work with
Biotech and life-science labs
Assay, sequencing and experiment data that needs models, pipelines and one shared knowledge base.
Research groups and universities
Analysis, prototypes and lab systems that outlive the person who wrote the first script.
Deep-tech startups
A senior team for the hard part of the product, before you can hire one.
Manufacturers and industrial teams
Vision, robotics software, sensors and AI agents on the equipment and systems you already have.
The matrix
Data and models
Years of data, still guessing what to test nextNeeds: Software, Models, Science
Each round of experiments is picked by intuition, while results from hundreds of earlier runs go unused.
What we build
A model trained on your past results that ranks the next candidates or conditions to try, and says how confident it is.
Who's on it
- Machine learning engineer
- Data scientist
- Computational biologist
Analysis that lives on one laptopNeeds: Software, Science
Results depend on who ran the script, and nobody can rerun last year's numbers.
What we build
A reproducible pipeline with versioned data, tests and a simple screen to run it.
Who's on it
- Bioinformatician
- ML infrastructure engineer
Data that isn't ready for AINeeds: Software, Models, Science
Years of records sit in old databases, lab systems and spreadsheets, and every question means a week of copying and cleaning.
What we build
Cleaned, structured and documented data with safe access, so models and agents give answers you can trust.
Who's on it
- Data engineer
- Data scientist
- Bioinformatician
Robotics, sensors and IoT
A robot that only works in the demoNeeds: Hardware, Software, Models
It picks perfectly in the lab, then fails on real parts, real lighting and real people.
What we build
Vision and control models trained on your own demonstrations, on the robot you already have, tested in simulation and on the real cell until they reach the agreed success rate.
Who's on it
- Robot learning engineer
- Perception engineer
- Controls engineer
Inspection is a person with a clipboardNeeds: Hardware, Software, Models
Defects slip through on busy shifts, and checking every part would take three more people.
What we build
A vision model on your existing cameras that checks every part, sends doubtful ones to a person and logs every decision.
Who's on it
- Perception engineer
- Machine learning engineer
- Full-stack engineer
Machines and processes that fail without warningNeeds: Hardware, Software, Models
You find out a batch or a motor went wrong when it's too late.
What we build
ESP32 sensor prototypes, a live dashboard and a model that warns of drift early and says why.
Who's on it
- Embedded engineer
- Data scientist
- Full-stack engineer
A prototype that won't become a productNeeds: Hardware, Software
The bench prototype works, but nobody knows how to make it reliable, updatable and safe.
What we build
Firmware, enclosure, cloud and testing taken to a version you can put in customers' hands.
Who's on it
- Embedded engineer
- Hardware design engineer
- Full-stack engineer
AI agents and code
Experts buried in repetitive paperworkNeeds: Software, Models
Quotes, intake forms and support replies eat the hours your best people should spend on real work.
What we build
AI agents linked to your tools that draft the work, with a person approving anything that matters.
Who's on it
- Agent engineer
- Full-stack engineer
Agents that can't safely reach your dataNeeds: Software, Models
Your team wants AI on your databases, code and documents, but nobody wants an agent holding the keys to everything.
What we build
Secure bridges (MCP servers) to each system, read-only by default, with permissions per agent and a full audit log.
Who's on it
- Agent engineer
- Full-stack engineer
- Project lead
Agents that work in the demo, not in productionNeeds: Software, Models
The prototype impressed everyone, then gave wrong answers to real users, and nobody could tell why.
What we build
Agent pipelines with evaluations, approval steps and monitoring, tested on your real cases before release.
Who's on it
- Agent engineer
- ML infrastructure engineer
AI-written code turning into a messNeeds: Software, Models
Features ship fast, but nobody understands the code, bugs keep coming back and every change breaks something.
What we build
An AI Code Audit, then the repairs: architecture, tests, security fixes and documentation your team can work with.
Who's on it
- AI code auditor
- Project lead
What we don't do
So nobody wastes a call.
- Tenders or bids that collect several quotes
- Grant or funding applications
- Selling or supplying equipment, instruments or robots
- Staff for secondment or extra headcount
How Noise Limited works
Every challenge gets its own iteration team, proposed around your problem, with a fixed plan from the brief to handover.
01About 2 minutes
Tech challenge brief
Tell us the problem in three short steps. We take the time to understand it and see whether our team can take it on, then arrange a first online meeting.
0215 minutes
Call and team proposal
A focused online call on your goal, data and constraints. We propose the iteration team for your problem, with the scope and price of the first iteration.
03From 1 week
Start the iteration
We work in Agile sprints. Each sprint has a fixed price, set by the size of the team and the scope, and agreed before it starts. Daily stand-ups in the team, and a review with you at the end of every sprint.
04Depends on scope
Build and hand over
Further iterations, each ending with something working you can test. Code, models, data and documentation are yours once paid for; keep the team on for support, or run it yourselves.
What we need from you
Access within a week
To the data, systems or hardware involved, under NDA.
One decision maker
Someone who can approve the first step and join the key calls.
About an hour a week
For check-ins, and to try what we demo at the end of each iteration.
Your team
A project lead who answers to you, the specialists your problem needs, and our own AI agents. You talk to the people doing the work, with nobody in between.
How we vet the bench
CVs, portfolios and even video interviews can be faked now. Every member passes five checks before joining a team, and keeps earning their place on every project.
- 01
Identity
Government ID with a live selfie check, a video call with simple live checks, and payment details in the same name.
- 02
Provenance
We confirm the work you show is yours: what you shipped, the code or commit history behind it, references checked through company email, and a walkthrough of your own past work.
- 03
Skill
A live, paid work sample in their field with AI tools allowed. We grade whether they check the AI's output, stay in scope and explain their trade-offs. Hardware engineers debug a real board; scientists analyse a dataset they have not seen.
- 04
Track record
Every project adds milestone reliability, client ratings and review by the Noise Limited lead.
- 05
Ongoing checks
Identity re-checked on each new project, random code reviews, and conflict-of-interest declarations.
Price and terms
Discovery Sprint
From US$2,400
Fixed fee. Credited in full against the build if you go ahead within 60 days.
- 2 weeks
- 1 project lead and 1 specialist
- Calls on Google Meet, about twice a week
- Written updates between calls
What you get at the end
- Milestone plan
- Each milestone with its acceptance criteria.
- Resource plan
- The team, hardware and data each milestone needs.
- Options analysis
- Two or three ways to build it, with cost, risk and time for each.
- Statement of work
- A fixed price for the first phase, ready to sign.
AI Code Audit
From US$2,400
Fixed fee. Credited in full against the repairs if you go ahead within 60 days.
- 2 weeks
- Project leads with our own audit agents
- Security, architecture and the risks in AI-written code
- A ranked list of fixes, each with effort and cost
Estimate your project
Every team has one project lead, plus the specialists your problem needs. Move the sliders to see a rough figure.
Your team
1 project lead, always included
Estimated total
From US$7,200
About US$1,200 a week
A rough guide, not a quote. The fixed price for each iteration is agreed after the Discovery Sprint, once we know the scope.
Then choose how to build
- Fixed scope
- A fixed price for each milestone, paid as each one is accepted.
- Ongoing support
- A monthly retainer for improvements, monitoring and fixes after launch.
One invoice per iteration
From Noise Limited, for the whole team, paid before the iteration starts. In US dollars, by bank transfer or card.
NDA first
Signed before anything confidential is shared. Every specialist signs one with us too.
Your data stays yours
Never used to train shared models, and kept in your region or on your own hardware on request.
Who owns what
- You own
- What we build for your project once it's paid for: your models, configurations, custom code and reports. Your data is always yours.
- Noise Limited keeps
- Our general tools, kits and know-how, which you can use free inside your solution.
- Never
- We never use your data, or models trained on it, for anyone else.
Building since 2012
- 2012Founded in Hong Kong
- The 2010sWeb and mobile systems for large companies
- Recent yearsMachine learning, AI agents and connected hardware
- TodayA deep-tech hub with a worldwide network
We started with web and mobile systems for companies including LG U+, Samsung and Sun Hung Kai Properties, and moved into machine learning, AI agents and connected hardware.
Questions clients ask
Do you sign an NDA?
Yes, before anything confidential is shared. Every member signs one with Noise Limited too.
How does a project start?
With a short, fixed-price first step: a two-week Discovery Sprint for a new problem, or a two-week AI Code Audit for existing code, from US$2,400. The fee is credited against the work that follows if you go ahead within 60 days.
Do you respond to tenders or write funding applications?
No. We work with one client on one problem from the start, beginning with a paid Discovery Sprint or AI Code Audit. We don't bid against other suppliers, and we don't write grant or funding applications.
Who owns the result?
You own what we build for your project once it's paid for, and your data is always yours. We keep only our general tools, which you can use free inside your solution.
Where is the team?
Remote, around the world. Noise Limited is registered in Hong Kong, and each team is picked for the problem and your time zone.
How do we pay?
One invoice from Noise Limited for the whole team, per iteration, paid before it starts. In US dollars, by bank transfer or card.
What is the smallest engagement?
A Discovery Sprint (two weeks, a project lead and a specialist) or an AI Code Audit (two weeks).
Tech challenge brief
Three short steps, about two minutes. We read every one to understand the problem before we arrange a first online meeting.
Press, partnerships or anything else? .