Zyra · The intelligent core
Zyra is the real-time machine-learning intelligence layer running inside ZenDMS. ZenDMS executes in the physical world of orders and logistics; Zyra is the layer that decides — without rigid, hand-written business rules.
Zyra isn't a product we're selling you. It's the clearest example of how we build: a real system, in live operations, solving a problem that rules engines had failed to solve.
01
The problem it solves
The decision path
Retailers run omnichannel order-to-delivery on legacy systems that handle ingestion, inventory mapping, and dispatch in silos, stitched together with manual handoffs. Each step optimises locally and no step sees the whole picture — so an order gets allocated to the warehouse with stock, then routed inefficiently, then dispatched to a carrier that misses the window.
Zyra ingests inbound orders from every sales channel and makes the execution decision in real time, balancing inventory availability, delivery proximity, and route efficiency simultaneously, then auto-dispatches through in-house or external logistics with no manual intervention. Embedded natively in the core ERP process layer, it optimises the full transaction from capture to drop, including automated carrier or field-agent selection.
02
How it's engineered
/01
Built on the IoTZen engine, not a single static optimisation model. Independent agents coordinate in parallel across inventory allocation (which warehouse or store), route efficiency, and courier or field-agent cost — recalculating and re-routing instantly when inventory shifts or a carrier misses an SLA.
Why it matters
A single optimisation model has to be retrained when the objective changes. Independent agents negotiating in parallel adapt when reality does.
/02
An open-source LLM fine-tuned on continuously captured master and transactional data from live enterprise deployments, with predefined business rules and constraints layered in. The model learns from real operational outcomes while hard business logic keeps every decision within safe bounds.
Why it matters
This is the architectural answer to the main objection to agentic systems. A pure LLM agent can do anything the model decides. Zyra's constraints sit outside the model, so no amount of model drift can breach them.
/03
The proprietary innovation sits in the orchestration and learning layer above the model: a self-improving semantic task-caching algorithm that learns an organisation's recurring workflows at runtime and reuses them — so decisions get faster and cheaper the longer Zyra runs.
Why it matters
Most agentic systems have flat or rising unit costs. This one has a declining cost curve, because a warehouse's Tuesday morning looks a lot like last Tuesday morning.
Live in production
Zyra is live inside ZenDMS, making allocation, routing, and dispatch decisions in production logistics operations. Model architecture and performance benchmarks are commercially confidential; we discuss them under NDA in a technical session.