You want to try something this week
Start with the Excel work — aggregation, manual re-keying, the monthly report. You ask in chat; the agent handles intake, aggregation and report generation. It ships as a product, not a project.
Ichiyomi Sheet → AI AGENTS
We build and roll out AI agents.
Manual re-keying, reconciliation, monthly reports. We build AI agents that take that work off your team — as a product you can start using now, as a setup that runs entirely inside your network, or as custom development against your requirements.
Talk to us before you have decided where to start.
Input Send the forms you already use Excel, PDF, scans, faxes, handwriting. Nothing about your formats has to change.
Agent Reading, checking, aggregating Every value is checked against the original. Only what cannot be grounded goes to a person.
Output Get it back in a usable form Aggregated Excel, dashboards, reports, even draft emails.
Three ways in.
Which one fits depends on how sensitive the data is, and how much has to be built to your requirements.
Some of your data cannot leave
Contracts or audits keep the data inside. We design and build a setup where all processing happens inside your network. Three weeks of verification fixes the scope, the price and the schedule.
On-premise →It has to be built to your requirements
Integration with your core systems, a workflow only you run, or shared drives holding drawings, specs and past projects — there, the scope of access and who is allowed to see what is the requirement. Three weeks in, you have a fixed price and schedule and decide whether to build.
Custom development →We build the error-catching before we hand anything to an AI.
"Roughly right" does not run a back office. This is where most of our implementation time goes.
Every value is checked against the original
Models misread values. That happens in any setup. We trace each extracted value back to where it came from in the source document, and anything we cannot ground never reaches the next step. This is what separates a demo from something that survives daily use.
Only the uncertain parts go to a person
If someone has to check everything, the automation bought you nothing. Human-in-the-loop review is reserved for the cases the agent is unsure about, so the checking itself shrinks.
The floor keeps working the way it does now
New input rules never stick. Your current forms and formats go in unchanged; we automate everything behind them.
We keep watching after go-live
Accuracy drops when a format changes or a new supplier appears. We monitor throughput and accuracy, and we stay through adoption so the setup keeps up with the work.
Want to know whether this fits your work?
Tell us the process and how the data is handled. We will be straight about what we can do and what we cannot.
Book a 30-minute call →Three weeks to a fixed scope, price and schedule.
- Day 0
30-minute call
The process, the data, your security requirements. If an existing tool is enough, or AI is the wrong answer, we will say so.
- Week 1–3
Verification (three weeks)
We test accuracy on your real data and settle the setup and the operating model. Remote-first: requirements definition, data handover and review all happen online. You get a quote and an implementation plan. Stopping here is a fine outcome. On-site visits can be arranged if you want them.
- Month 2 –
Rollout and trial run
One process, one team first. Integration with existing systems is confirmed at this stage.
- After go-live
Production and accuracy monitoring
We widen the scope while monitoring accuracy and throughput, and follow your format changes as they come.
What people ask before the first call
We do not know where to start.
That is fine. In a 30-minute call we look at where your time actually goes, and separate what an AI agent can take over from what it cannot. Picking one process to start with works better than picking five.
Can we start small?
Yes, and we recommend it. One process, one team; confirm the result, then widen the scope. Opening with a company-wide requirements exercise is the most reliable way to stall.
Do you work in English?
Yes. Projects run in English end to end — requirements, delivery, documentation and support. Delivery is remote-first: requirements definition, data handover and review all happen online, so distance is not a constraint. On-site visits can be arranged when a project calls for it.
Will our data be used for training?
No. Data you hand us is never used to train models, and the rights stay with you. If contracts or audits mean the data cannot leave at all, we propose a setup that runs entirely inside your network.
What does it cost?
It depends on the shape of the engagement. The cloud product is priced per user per month with a minimum number of seats; building inside a closed network and custom development are quoted per project. Either way, three weeks of verification fixes the total and the schedule before you commit — we will put concrete numbers in front of you on the first call.
Start with "which process should we take on first?"
You do not need settled requirements. Tell us the situation and we will propose a way in.
A 30-minute call
Tell us the process and how the data is handled, and we will work out what is buildable while we are on the call. If it will not work, we will say so.
Book a call →You would rather see the product first
If the work lives in spreadsheets, the cloud product may be enough on its own. Start there and see.
Ichiyomi Sheet →Cameras and sensors on the floor
Work on images and video on the floor is a separate line — edge AI.
Tinyboom →