vyntra AI LABS · A RAPPORTSOFT VENTURE · PUNE — EST. 2025
Applied AI for logistics infrastructure — gates, yards and weighbridges that see, decide and sync themselves. Sixteen years of field DNA from our parent, one office in Pune, clients in eleven countries.
GATE AUTOMATION · ANPR · ACCESS CONTROL
The gate that admits trucks itself. Number plates read at 99.3% — monsoon rain included — e-way bills and gate passes verified on the spot, barriers commanded without a clipboard. A decision in under 200 ms, an audit trail that outlives the shift.
YARD MANAGEMENT · SLOTTING · TRAILER RE-ID
The yard that parks itself. Trailers re-identified at 97.1% from fixed cameras, slots assigned by arrival forecast, dwell-time risk flagged before it costs you a demurrage hour — one live map of every wheel, dock and door.
WEIGHBRIDGE · SLIP OCR · AUDIT CHAIN
The bridge that weighs without slips. Capture gated to the vehicle, handwriting on weighment slips read by OCR, every weight hashed and chained — tamper-evident by default, synced to ERP while the truck is still rolling off the plate.
01 — ABOUT US
vyntra — from Sanskrit yantra, “instrument.” We are not a startup that discovered logistics last year: our parent, Rapportsoft, has been building systems for plant floors since 2009. In 2023 an internal team pointed cameras at factory gates. In 2025, that team became vyntra AI Labs.
Today our instruments run at 300+ client sites across eleven countries — admitting trucks, parking trailers and weighing cargo without a human in the loop. The through-line from parent to lab is the same: physical operations generate enormous data that nobody uses. We build the instruments that use it.
We are 48 engineers in one office in Pune — vision researchers, ML engineers, edge hackers and support engineers who answer at 3 a.m. because a gate in Rotterdam is down and a truck is waiting.
02 — WHAT WE DO
Three disciplines and one habit: models that hold up outdoors — in dust, rain and shift changes.
AI at vyntra means systems that decide — not chatbots. When a truck reaches a gate, our policy engine weighs the gate pass, the e-way bill, the vehicle’s history and the yard’s current state, then admits, diverts or escalates it in under 200 ms.
The hard part isn’t the model; it’s the boundary. We encode what must be certain as rules, learn what can be probabilistic, and keep a human in the loop exactly where regulation and common sense demand one.
IN PRACTICE
Forecasts and anomaly scores that actually run the operation: appointment ETAs, dwell-time risk, weighment fraud scores. We stay classical where classical wins — boosted trees beat transformers on tabular ops data, and we say so out loud.
Every score ships with its explanation. Site supervisors don’t trust numbers they can’t argue with, so we show the driving features — in their language, not ours.
IN PRACTICE
Our models read the physical world: plates in monsoon rain at 40 km/h, seal integrity at dusk, handwriting on a weighment slip. This is the lab’s core — a parent company’s sixteen years of site access, distilled into footage, labelled twice.
Then we shrink it. Distillation and quantisation until a Jetson at the gate does in 60 ms what a data centre does in 500.
IN PRACTICE
A model that needs the cloud is a model that fails when the internet does. Ours run at the gate, on the bridge, in the yard — with telemetry, shadow deployments and canary rollouts like any grown-up software.
We watch accuracy the way operations watches uptime: nightly evals, drift alerts, and a rollback path that takes minutes, not meetings.
IN PRACTICE
03 — HOW WE WORK
A dashboard tells you yesterday. An instrument decides the next truck. Everything we ship has to close a loop, not just draw one.
The yard does not stop when the internet does. Every decision runs on-device and syncs when the link returns — no exceptions, no spinners.
Every score carries its reasons, in the operator’s language. Supervisors don’t trust numbers they can’t argue with — neither do we.
If a rule solves it, we don’t train a model. If a model solves it, we don’t build a platform. Boring where it counts, sharp where it matters.
Every event is hashed, chained and auditable. The register book died in 2009; the proof stayed.
04 — START A PROJECT
One call to scope it, one site walk to instrument it, one pilot to prove it — gates, yards and weighbridges that see, decide and sync themselves.