Vision
Brain exists because the most expensive work in the VinMake studio is also the most repetitive: reading a client tech pack, drafting a BOM, calculating true cost, producing a quotation, drafting a polite reply. A merchandiser like Mai spends hours on each new RFQ. A manager like Zean spends hours trying to figure out which inquiries have gone cold. We’re building the system that turns that work into a fast review loop instead of a slow drafting loop. AI writes the first draft. A human reviews, corrects, and approves. The corrections feed back into the system so it gets better at drafting next time.The bet
The bet is that in garment operations, a controlled AI assistant beats both a general chatbot and a fully automated system. A general chatbot doesn’t know which workflow you’re in, where data should land, or whether a draft is safe to send. It’s a clever toy. It doesn’t change the day. A fully automated system writes records and sends emails without a human in the loop. The first time it gets a button count wrong, sends the wrong file, or quotes the wrong margin, the team loses trust and stops using it. A controlled assistant does only the things we’ve defined, drafts what’s drafting work, asks for approval on what matters, and links every output back to its source. It earns trust by being predictable and being useful at the same time. That’s the product.What “AI-first draft, human approval” means in practice
It is the spine of every workflow in Brain.- AI prepares the first version: extracted fields, draft BOM, draft true cost, draft quotation, draft email.
- Human reviews and edits in a clear UI that shows confidence and source.
- The system commits only after human approval, and only through the approved write path (CutMake for production records, email send for client comms).
- Every correction is logged: AI draft vs. human final, what changed, why. This is the feedback loop that makes the system better over time.
Why iterative, not big-bang
The system gets built one workflow at a time, deployed to real users (Mai and Zean) on real data, then iterated. Each shipped workflow has tracing, correction logging, and a changelog entry. Nothing ships without those three. The reason is simple. A monolithic launch with eighteen features hits the day-one trust problem head-on. One workflow that genuinely saves Mai an hour is worth more than eighteen features she has to babysit.What Brain is not
- Not a chatbot. No general “ask me anything.” Every entry point is a defined workflow or query template.
- Not a CutMake replacement. CutMake stays as the structured source of truth. Brain handles the work between an RFQ arriving and a record existing in CutMake.
- Not an autonomous agent. Brain doesn’t send emails, finalize quotations, or commit important records without a human approval gate.
- Not a research tool. Brain doesn’t browse the web for unrelated questions. Pricing intelligence is a defined workflow with bounded data sources.
The shape of success
We’ll know Brain is working when:- Mai chooses to open Brain instead of doing it manually, every time.
- Zean can ask “what’s happening with Jaspal style J26WGW153?” and get a complete, source-linked answer in seconds instead of digging through email threads.
- The correction log shows the system’s first-draft accuracy improving month over month on a measurable subset of fields.
- No RFQ goes cold because nobody noticed.
- The team trusts the drafts enough to review fast, not rewrite from scratch.