VAUNT.Shinjuku
Tokyo
Free consultation

WORK

What changed,
and how it was built.

Better-run data collection, a research environment, automated hardware testing, and open-source development with AI. The problem, the design and what changed afterwards.

The client work below is our founder’s own experience, including work before the company was established. Client names and confidential details are withheld; each case describes the founder’s own responsibilities.

01 / DATA & RELIABILITY

Founder’s work

From daily manual recovery
to a pipeline you can trace.

API AAPI BAPI C QUEUESplits the work WORKERRuns the job Records state, recovers from failure
A conceptual diagram of the system. Not a copy of actual screens or architecture.

Python / DRF, FastAPI / Cloud Tasks / task state tracking

Digging through logs and recovering by hand, every day.

Every time data collection stopped, a developer had to search the logs and run recovery commands. Error output was sparse, and maintenance was eating into time for new development.

Start with alerts, then change the structure.

Alerts came first. The serial processing was then rebuilt on a worker–dispatcher model that runs jobs through a queue. Tasks were given states such as waiting, running, succeeded and failed, and a recovery endpoint that is called automatically was added.

From one API to several applications.

The shared design was rolled out to applications for several APIs. Another department that needed the same API asked to use one of them, source code and all, and received the design, shared code and tooling.

WHAT CHANGED

A system that tracks state and recovers on its own. Data rarely has to be fetched again, and the design was left in a form others could reuse.

Discuss data integration or system improvement

02 / CLOUD & RESEARCH

Founder’s work

A research environment
shaped by the data it holds.

Separate accounts and network paths ACCOUNT / AACCOUNT / B Research app Data & analysis
A conceptual diagram of the system. Not a copy of actual screens or architecture.

AWS / Transit Gateway / SageMaker / Flask / React

A usable environment for research on personal data.

The research environment spanned several AWS accounts, and its network rules allowed no traffic to or from the public side. The design had to make the necessary research workloads possible while also meeting conditions specific to SageMaker.

Design it, validate it, then build it.

Our founder worked out the network segments and traffic paths, confirmed feasibility with AWS’s own validation tools, and carried the work through to proposing and building the environment with Transit Gateway and other services.

Right through to the applications on top.

The work also included a database-backed research support application in Flask and React, covering features researchers rely on such as uploading and downloading data and running analyses.

WHAT CHANGED

Working within the network and data handling requirements, our founder delivered both the research infrastructure and the application running on it.

About app and cloud development

03 / ENGINEERING & AUTOMATION

Founder’s work

Thousands of tests, automated
from execution to verdict.

CSV PythonTCP / run and judgeMany cases checked in one run
A conceptual diagram of the system. Not a copy of actual screens or architecture.

Python / TCP / CSV / hardware testing

Type a command, read the log, repeat.

Testing video transmission and switching equipment meant powering up the unit, typing commands into a terminal by hand and reading the logs. With so many combinations of signal switching and aggregation, the test cases ran into the thousands.

Build a list of operations and a way to run and judge them.

The test commands were organized into CSV files, and a Python script sends them one by one over TCP. It automates not only running the commands but also judging each result pass or fail.

Thousands

Hardware tests run in bulk and judged automatically.
The number of test cases our founder was responsible for.

WHAT CHANGED

Typing and checking that people used to repeat by hand became a system that runs and judges tests in bulk.

About automating repetitive work

04 / AI-ASSISTED DEVELOPMENT

Founder’s open source

furnace_

Building software
together with AI.

An inference server and library written in Rust.
It lets applications put AI models to use.

THE SOFTWARE

From input to inference,
delivered as a service.

It accepts inference requests over HTTP and gRPC, and the model implementation can be swapped out. A Rust project that also handles load control and processing metrics.

INPUTMODEL RUST / HTTP, gRPC / swappable models

A conceptual diagram of what the software does.

THE WAY WE BUILD

Start from the problem,
then build and verify, again and again.

  1. 01

    Frame the problem and the options

    Connect a real market problem with our own skills, and weigh the options from business, design and other angles.

  2. 02

    Build with AI, review in a separate role

    Using Claude and ChatGPT, with implementation and a mentor role kept apart, cycling through findings, fixes and merging changes.

  3. 03

    Verify the complex flows

    Model the mechanism and verify specific properties, such as that processing terminates, concurrency limits hold and deadlines are handled.

About the technical verification

furnace includes TLA+ specifications and model-checking records for how work is dispatched, routed and given execution slots. Formal verification methods from our founder’s university research are applied to design checks in day-to-day development.

Verification applies to what was modeled and the assumptions made. It is combined with testing and review of the implementation to assess the software as a whole.

Using AI in development, carried through to design and verification.
We bring the same approach to the service you want to build.

Discuss AI-assisted development

KEEP EXPLORING

Taking the technology to the next problem.

Concept stage

Making government documents easy to follow.

We are exploring a guide that rewrites explanations of public programs and procedures in plain Japanese so people can see what they need to do. Its benefit to users has yet to be tested.

Exchange ideas with us
Interactive experience

Set the flow free.

Release 1,024 tasks at once, steer parallelism and recovery, and experience the machinery behind the screen as flowing light: FLOW FIELD.

Try the playground

LET’S TALK.

YOUR NEXT CHAPTER.

No spec yet?
That’s fine.

Tell us what’s slowing you down or what you want to build.
That’s where we start.

Automation & AISpend less time re-keying and checking.Product development & system workBuild something new, or make what you have better.PartnershipLooking for an engineering partner for your clients?

The first consultation is free. We’ll review what you send and tell you whether we can help and how we’d go about it.
Detailed research, design and prototyping start only once we’ve agreed on scope and cost.