Proposed by the AI
Decided by a person
BriefingDrafted by the AI
Plant expansion, phase two
A phased expansion of the production block. Site visit done. The client wants the schedule before the next budget tranche.
Next: send the schedule before Friday.
Set as next step
ReplyDrafted by the AI
Re: phase two, the schedule
Thank you for Tuesday’s site visit. As promised, the phased schedule for phase two is attached, ahead of Friday’s review.
I have read thisSend
QuotationDrafted by the AI
Phase two, from the email thread
Design and engineering; supply and installation; commissioning.
Total: about ₹1.1 Cr
Approve
Quotation · Plant expansion, phase two
Further to the site visit, we are pleased to quote for phase two at about ₹1.1 CrHeld for code₹1,10,40,000.
Written by the AIComputed in code
- Design and engineering₹18,40,000
- Supply and installation₹79,60,000
- Commissioning₹12,40,000
- Total
Worked out again in the database: it agrees. A total that disagrees is refused.
Our code
- FastToday’s modelA better oneSwapped
- SmartToday’s modelA better oneSwapped
- VisionToday’s modelA better oneSwapped
- Briefing · setUnchanged
- Reply · sentUnchanged
- Total · ₹1,10,40,000Unchanged
AI systems for the enterprise07 / 09
Enterprise AI
The AI proposes. Your people decide.
What we doARK
/ 09 · Be built around
Enterprise AI systems for India’s leading companies
We don’t bolt a chatbot to the side of your company. We build AI into the work itself: it reads, briefs and drafts; a person says yes before anything leaves; and every number is computed in code the AI cannot bend.
01 · Three drafts
The AI drafts. A person taps.
The AI drafts all three. None is a decision until someone who answers for it says yes.
02 · The number
The words are the AI’s. The number never is.
Code adds up the lines. The database adds them again, and refuses a total that disagrees.
03 · The models
Change the model. Nothing else moves.
Every model sits behind our own code. A better one goes in, and the work carries on.
An illustration on demonstration data: the deal, its words and its figures belong to no client.
Who it is for
AI in the work. Numbers out of its hands.
Enterprise AI at ARK is for the company that wants AI doing real work, reading enquiries, briefing deals, drafting the reply and the quotation, without handing it a decision or a figure. The AI proposes. Your people decide. Every number a customer or a board will read is computed in code.
We don’t sell a chatbot, a demo or a wrapper around one vendor. We build the system your people use every day, and we stay until it runs.
Three kinds
What we mean by enterprise AI.
Assistants.Copilots.Pipelines.
Assistants, copilots and pipelines built into how your work already moves. An assistant that turns a voice note into a task with an exact time. A copilot that briefs every deal before the rep calls. A pipeline that reads a visiting card photographed at an exhibition and turns it into a lead, ready for the rep to confirm.
- If the job is the work between the work, routed and sent on time, that is workflow automation.
- If the system does not exist yet, that is custom software.
- Not sure where AI would pay first? Start with a gap analysis.
Enterprise AI is one of nine services built to work as one system.
Three rules
Built to be trusted.
Three rules run through every system we build. The AI drafts: briefings, replies, quotations, the prose of a report. A person decides: nothing is sent, merged or changed without a yes. Code computes every number: targets, forecasts, totals and times, worked out where the AI cannot reach them.
| The job | The AI drafts | A person decides | Code computes |
|---|---|---|---|
| The briefing on every deal | Three sentences and a next step | Sets the next step | — |
| A reply to a client | Drafts it | Reads it, ticks it, sends it | — |
| A quotation | Writes the words and the lines | Decides it goes | Computes the money; the database checks it again |
| A lead’s temperature | Proposes hot, warm or cold | Applies it | Accepts only those three |
| Targets, forecast, attainment | — | — | Every figure, every time |
| A reminder at the minute | Reads what you said | — | Works out the minute; a clock with no AI in it keeps it |
| Anything that leaves the company | Proposes it | One tap | No path lets the model send |
01 · In production
Intellreach: the CRM that reads the pipeline.
Our AI-native CRM, in production since September 2026. Every enquiry from seven channels lands in one record; the AI briefs every deal and drafts the reply, and code computes every number.
Its AI
- Briefs every deal in three sentences, with a next step
- Drafts replies, and quotations from the email thread
- Reads visiting cards and project listings from a picture
- Writes the prose of every report
Its code, and its people
- Targets, forecasts and attainment, computed in code
- Nothing sent, merged or changed without a person’s yes
- Who may see what, decided by the database
- Nothing deleted; every change audited
02 · In daily use
ARK Assistant: say it once, it gets done.
An executive assistant for every desk, in daily use since August 2026. The AI reads what you say; a clock with no AI in it keeps the time. 2,480 automated tests stand behind it, 253 of them on the reminder path alone.
Its AI
- Hears the voice note, and shows what it heard
- Reads it with what it knows about you
- Files the task under the right client
- Phrases your briefing at eight every morning
Its code, and its people
- Turns “tomorrow 4pm” into an exact minute, in Indian time
- A clock with no AI in it checks every fifteen seconds
- Below a confidence bar: “I don’t have a clear record of that”
- Nothing leaves without your tap
One gate, six models
How we choose models.
We choose a model for each job, not a vendor for the company: a fast one for short, frequent work, a stronger one for the prose of a report, a vision model for a photograph of a visiting card. Before a model goes near your people, it goes through a gate of tests; for ARK Assistant, six models went through the same gate, and one was kept.
Every model sits behind a wall of our own code. It reaches your data only through that code, it never decides on its own, and it never computes a number. So when a better or cheaper model arrives, it goes in behind the wall and nothing your people rely on changes. You own a system that keeps working as the models change.
Questions
What leaders ask before they trust it.
- What do you mean by enterprise AI?
- Assistants, copilots and pipelines built into how your work already moves, not a chatbot bolted to the side of it. In Intellreach, our AI-native CRM, the AI reads every enquiry, briefs each deal and drafts the reply and the quotation; the people decide what goes out; and every number in every report is computed in code the AI cannot bend.
- How do you make enterprise AI reliable enough to run on?
- By keeping the AI away from the things that must never fail. ARK Assistant turns what you say into timed tasks with a language model, then hands the time to a clock with no AI in it, which has delivered 850 reminders in five weeks of daily use without missing one. Behind it sit 2,480 automated tests, 253 of them on the reminder path alone.
- Will it work with the systems we already run?
- Yes, by design. The brain is kept separate from its faces, so the same core answers in Telegram, in a web app or inside another system, and we connect it to what your company runs as part of the build. Every action the AI proposes waits for a person’s tap before it leaves.
- Which AI models do you use?
- The one each task needs, chosen for the job and swapped when a better one comes, because every model sits behind our own code and never touches your data directly or decides on its own. You buy a system that keeps working as the models change, not a wrapper around one vendor.
AI your people can put their name to. It drafts the work, they decide it, and code keeps every number. Tell us where the work waits on a person, and we’ll show you what the AI should draft.
Workflow automation08 / 09
Automations
The follow-up nobody owned, owned by a system.
What we doARK
Keep scrolling
to the next service

