Tali Treasury
AI-Powered Payroll, Expenses & On-Chain Protection
Created on 5th September 2026
•
Tali Treasury
AI-Powered Payroll, Expenses & On-Chain Protection
The problem Tali Treasury solves
Payroll, employee requests, statutory contributions and expense reimbursement are often handled across separate systems. This creates repetitive data entry, slow approvals, limited visibility and a dangerous gap between approving a payment and actually completing it.
Tali brings these workflows together. Employees can submit overtime, leave and receipt claims through one mobile-friendly application. Employers can review each request, understand its payroll impact, run payroll and monitor team earnings. Treasurers can manage expense limits, review exceptions and track reimbursements.
For payroll, Tali calculates employee pay together with EPF, SOCSO and EIS allocations. Approved overtime and unpaid leave are incorporated into the payable period before the final calculation. Salary and statutory payments are then executed atomically through a Sui Move contract. If any required contribution is below the mandate’s minimum, the entire transaction is rejected and no USDC moves.
For expenses, Gemini extracts structured information from receipts and timesheets, reducing manual entry. Deterministic policy checks evaluate limits, categories, dates, duplicate receipts and approved recipients. Uncertain claims are sent to a human reviewer instead of allowing AI to control funds.
Tali also supports salary streams, allowing employees to see earnings accrue over time and withdraw vested pay using their own wallet. Every confirmed payment or refusal can be independently verified through its Sui transaction.
Challenges we ran into
Malaysian statutory payroll rules do not treat every payment component equally. EPF uses contribution schedules and excludes overtime from its wage basis, while SOCSO and EIS include overtime and apply wage ceilings. Unpaid leave must also reduce the wage before the statutory split is calculated. We separated each statutory calculation, encoded the correct calculation order and added extensive tests for wage bands, ceilings, overtime and leave adjustments.
Blockchain payment failures can be ambiguous. A network timeout does not necessarily mean that a transaction failed, and automatically retrying could create a duplicate payment. We solved this by deriving and storing the transaction digest before broadcasting, using atomic compare-and-set state transitions and reconciling the exact digest on demand. An uncertain payment is never automatically signed or submitted again.
Wallet connection alone was not secure enough for authentication. We implemented signed, single-use wallet challenges, one-hour HTTP-only sessions, exact-origin validation and account-switch invalidation. Receipt analyses are stored as private, short-lived drafts that can be consumed only once when creating a claim.
We also needed to make complex blockchain and payroll evidence understandable on mobile screens. Early versions contained dense calculations and long technical explanations. We redesigned the interface around clear outcomes, compact summaries and expandable evidence, allowing users to understand the result first while still giving judges access to policy checks, exchange-rate details and transaction proofs.
Tracks Applied (2)
PAYMENTS & STABLECOINS
Sui Foundation
AI × SUI
Sui Foundation
Technologies used