We designed and built Velora AI Hub for a regulated brokerage. Staff store and share documents in one place, find them by meaning, get AI summaries and ask an assistant. Next, team chat keeps shared files inside the same rules.
Delivery
37 daysfrom first commit to production00Project overview
Velora Markets is an online FX/CFD brokerage operating in several Asian markets. Codech took the project from requirements to a live system: design, prototype, build, deployment and operations. Phase 1 went live in 37 days.
The challenge
KYC files, client agreements, trade records, compliance policies and research were spread across personal drives and email. Search only matched filenames, scans and Chinese documents were invisible, and a regulated firm needed every change on record.
Our proposal
We wrote an interactive proposal that mapped the workflow and the AI design, then built a clickable prototype the client approved screen by screen. Phase 1 focused on the document workspace, and Team Chat was scoped as a later phase so the core could ship fast.
The solution
A workspace that works like Google Drive, running in the firm's own cloud. AI reads every upload so staff can find files by meaning, in English and Chinese.
01Integrations
Staff sign in with the Microsoft accounts they already have. Documents stay in the firm's own cloud in Singapore, and the AI runs behind one permission check.
Sign-in & AI02
Data & search02
Cloud & delivery02
02The problem
KYC files, client agreements, trade records and policies lived in personal drives and email threads. Nobody had the full picture.
Scanned PDFs and Chinese documents were effectively invisible. If you didn't know the file name, you didn't find the file.
A file sent in Teams is a detached copy: no classification, no version history, no access log, and no way to take it back.
03Document Inventory Live
Shared drives that work like the tools staff already know, with AI built into every file: it reads, summarises and answers, but only from documents the person is allowed to see.
Shared drives
Personal, team and company-wide drives with folders, grid and list views, stars and recents. Every file shows its classification, and scanned images are read with OCR so they're searchable too.
AI search
Keyword and semantic search run together, so "passport expiring soon" finds the KYC pack that never uses those words. Works on scanned PDFs, images and Chinese text.
AI summaries
Each file gets a summary, category, tags and key facts such as client, owner and expiry dates, so you know what it is before you open it.
AI assistant
The assistant finds and reads documents for you and cites what it used. It can suggest changes, like a new folder, but nothing happens until a person confirms.
AI actions
Open a document and pick an action: summarise it, pull out key facts, check unresolved comments, compare it with another version, or ask where it should be filed.
04Team Chat In build
Channels and direct messages inside the same workspace. When someone shares a file, the message points to the real document, so it keeps its classification, permissions and latest version.
Channels & DMs
Messages carry a live link to the hub file, not a copy. Rename it and every message updates. If someone can't open it, they see a locked card instead of the contents.
Quick Send
Send several files or whole folders from any screen. If some people in the channel can't open them, you're told who, and you can grant access on purpose rather than by accident.
@ai in any channel
Type @ai and the assistant answers for the whole channel, citing the documents it used. In chat it can only read, never change anything.
The key design decision
A channel grants no document access. Ever.
The shortcut was to reuse existing user groups as chat channels and get file sharing for free. We rejected it. Groups can be given whole drives, so adding someone to #deals would silently hand them every document ever posted there.
Instead, opening a file from chat runs the normal permission check against the document itself. Sharing stays the only way to grant access, so adding chat doesn't add a single new way for documents to leak.
05How it works
Uploads never wait on the AI. A background worker picks each file up, reads it and makes it useful everywhere in the workspace.
06How we delivered
We pitched with a working proposal site. The client then clicked through a full prototype before we wrote any production code, and that approved prototype became the contract for every screen. Both are below: scroll and click around.
An interactive proposal in English and Chinese: the modules, the architecture and the delivery plan.
Tell us about one workflow you'd like AI to handle. We'll send you a proposal like this one, with the AI design, the screens and the plan, plus an interactive prototype you can click through. Free.
The client approved it screen by screen. Every production screen is screenshot-tested against it.
The real files from the engagement. The prototype uses demo data and includes the Team Chat screens.
Specs, then a clickable prototype of every screen that the client reviewed and approved.
Each production screen is screenshot-tested against the approved design. More than 0.1% drift fails the build.
Staging in week 4, production on day 37, and new modules like Team Chat added in slices.
07Under the hood
It can only propose actions; a person confirms each one. Sharing and downloading are outside what it can do at all, because document text is treated as untrusted input.
A 26 MB download took 14 attempts from Beijing. Resumable downloads fixed it, and an accelerated route measured 58% more throughput from China.
Every change writes an audit entry in the same database transaction, and the log is append-only at the database level, not just in code.
We'll look at where your documents and conversations live today and show you what an AI workspace would change, in a free 45-minute call. Then we send you a free AI proposal and an interactive prototype.