A bilingual AI assistant that turns Oracle EBS HRMS into a conversation — leave, certificates, attendance, approvals — for around 10,000 employees, hosted inside the Kingdom.
How HexaFlow built it: one conversation for leave, certificates, attendance and approvals — with Oracle EBS HRMS still the system of record behind every answer.
The HR platform was not the problem. Oracle EBS HRMS held employee records, organisational structure, leave, attendance and approvals correctly for around 10,000 people. Reaching it was the problem: employees moved between self-service portals and separate applications to do things that take one sentence to describe.
That fragmentation lands on the HR team as workload. A large share of their week went on repetitive inquiries — how much leave do I have left, where has my certificate got to — and on manually chasing the status of requests already in the system.
Availability was the third constraint. HR services effectively existed during office hours on a desktop, while the workforce is mobile and distributed. An employee on a site at 9pm had no route to a payslip or a leave balance.
And the answer could not be a second system of record. Anything that held its own copy of leave balances would drift from Oracle within a week and become the thing nobody trusts.
The answerHR services delivered in the app employees already have open, with Oracle EBS still the system of record and every request auditable end to end.
WhatsApp rather than another portal or another app to install. Adoption is the whole point of the project, and an interface with nothing to learn and nothing to download is the only one that reaches a workforce of this size.
Natural-language understanding maps a message in Arabic or English onto an HR service, extracts the parameters it needs, and identifies the employee and their reporting line from Oracle — so the assistant knows who is asking and who has to approve before it asks anything back.
Pick any of the fifteen services. Messages are illustrative.
Balances, attendance records, profile and organisational data are read live from Oracle EBS, and submissions are written straight back into it. Oracle remains the authoritative record; the assistant is an interface to it, never a parallel truth.
That was a condition, not a preference. Anything holding its own copy of leave balances would drift from Oracle within a week — and become the thing nobody trusts.
An interface to the record, never a parallel truth. What it keeps is the work around a request:
A request raises the real workflow — direct manager, senior manager on escalation, HR officer — with each approver acting from their own WhatsApp thread and the employee notified as the request moves. Status tracking stops being a phone call.
“I’d like to request leave next week.”
You are each approver in turn. Approve, or escalate from the direct manager.
Salary and employment certificates are generated from Oracle data on approval and delivered into the same conversation, rather than produced by hand and emailed later.
Arabic and English run all the way through: in what the assistant understands, in how it replies, and in the documents it generates — which are formatted right-to-left when they are in Arabic.
Sample record. Pay values masked.
To whom it may concern,
This is to certify that Layla Hassan, employee number 104872, is employed as Site engineer in the Operations department, and has been with the organisation since 2019-03-12.
Their basic salary is •••••• SAR, with allowances of •••••• SAR.
This certificate is issued at the employee’s request.
Generate it, or switch the certificate or its language. Sample record; pay masked.
The platform runs on public cloud inside Saudi Arabia, with employee and HR data resident in-Kingdom, logical isolation of tenant data and processing, and an AI environment scoped to the organisation — the condition on which everything else depended.
Public cloud hosting inside the Kingdom. Employee and HR data stays in-Kingdom.
Logical isolation of the organisation's data and processing, and an AI environment scoped to it.
Enforced server-side against Oracle's own hierarchy.
Every request keeps the message that raised it, every approval, and the Oracle transaction it produced.
WhatsApp messages arrive at a conversation gateway, are classified and parameterised by the language layer, and resolve against an employee identity taken from Oracle. A workflow engine holds request state and the approval chain, while an integration layer performs the reads and writes against Oracle EBS. Everything runs in-Kingdom.
Built to extend. The service catalogue is designed so new HR services — and later, functions beyond HR — onboard without redesigning the platform.
HR services reachable in the app employees already have open, in Arabic or English
Oracle EBS kept as the single system of record — no second copy of leave balances to drift
Approvals run end to end in the conversation, with the employee notified at each step
Repetitive balance and status inquiries taken off the HR team
Available outside office hours, from a phone, without a desktop portal
Employee and HR data resident in Saudi Arabia, with isolated processing
A service catalogue built to extend beyond HR without re-architecting
It calls every customer, then tells the board what they said
An AI agent calls every customer after service and gathers the review on the call itself. Every conversation is then analysed automatically and clustered into a picture of what is happening per department, per individual and across the whole organisation.
Three modules that read a feasibility study, judge it, and write a new one from scratch
Three modules: one reads a feasibility and fills the whole Excel model, one benchmarks it against world standards and returns a gap analysis, and one generates an entirely new professional feasibility from historical data.
Construction takeoff and bidding from architectural PDFs, in minutes
Reads architectural PDFs, detects the scale on its own, measures only the trade you work in, and returns quantities, labour hours and a bid — then pushes the work into your CRM.