Built so the obvious mistakes cannot happen.
The claim is not that the system never makes a mistake. The claim is that the failure modes a tired human makes at 2am are structurally prevented, and that everything else is escalated instead of guessed.
Accuracy by construction.
Two independent signals have to agree before anything is approved, and the pools those signals come from never mix.
- Pools stay separate. UPI is never cross-matched against IMPS or NEFT, even at an identical amount.
- Nothing is guessed. Ambiguous or reused entries are escalated with their history.
- Timeout clocks are anchored to stored timestamps, so a restart loses no in-flight state.
- Every loop iteration is guarded. One bad entry is skipped and the pipeline continues.
The two-key rule
Amount alone is never sufficient. That single rule removes the largest class of mismatches in deposit reconciliation.
Where your data lives.
Deposit data does not leave your infrastructure. There is no third party sitting in the approval path.
Private deployment
The application, the database and the browser all run on your own server. Deposit data does not leave your infrastructure.
Isolated browser profiles
One persistent profile per site, kept outside the application directory so a deploy can never disturb a live login.
Credential hygiene
Secrets live in environment configuration and are never committed to source. Access can be rotated without a code change.
Timezone correctness
Every timestamp and day boundary is Indian Standard Time, enforced at the data layer, so today always means today in IST.
Full audit trail
Every state change is attributable to an actor: automation, a team member on the platform, or a team member on the dashboard.
Nothing guessed
Ambiguity is escalated, never auto-resolved on partial evidence. There is no path where a weak signal becomes an approval.
What keeps it up at 3am.
A 24 by 7 operation is only credible if the system can repair itself while nobody is watching.
Watchdog every minute
Checks the display stack, the browser, the web app and all three workers, and repairs whatever it finds down.
Restart with backoff
Failures restart automatically, with exponential backoff and a restart ceiling so a crash cannot become a loop.
Memory cycled early
Both Node and the browser are watched and cycled before memory pressure becomes visible to a client.
Circuit breakers
Repeated gateway failures trip a cooldown, keeping the machine responsive while a page or session is repaired.
Timers survive restarts
Timeouts and holds are anchored to stored timestamps, so a restart never loses in-flight state or resets a clock.
Graceful shutdown
Workers finish the current iteration and save state cleanly rather than being killed mid-write.
Four layers, one process on your own server.
Nothing in the approval path depends on a third party service. The browser, the engine and the database all live on infrastructure you control.
Sources
- Google Pay (UPI)
- Bank portals (IMPS, NEFT, UPI)
- Trading platform
Capture
- One persistent browser
- Isolated session per site
- Parallel tab pool
Engine
- Matching loop
- Decision rules
- Audit writer
Surface
- Live dashboard
- Ledger and P/L
- Alerts and notifications
Watch it clear your queue.
Bring a day of real deposit activity to the demo. We will run it through the engine and show you exactly what it resolves on its own.