Operational twins
A live, governed mirror of a real object, such as a farm plot fed by satellite and soil sensors, or a tower fed by network telemetry.
One method for any object,living or non-living.
ONE CONNECTED INTELLIGENCE
01 / THE AGENT NETWORK
Most twins are built one asset class at a time. We start one level higher, with a taxonomy that can describe anything — so the same platform can model a wealth client, a field service job and a fibre route.
SignalsConsented internal, ecosystem and external data.
TwinThe governed model of the object is updated.
AgentsSpecialist agents simulate, predict and draft.
GroundingRetrieval from approved, cited sources only.
ReviewA named owner approves, edits or rejects.
Action and learningThe outcome is recorded and feeds back.
01 — Digital twins
Three ways to twin.
Once an object is defined, we can twin it in three ways.
Live mirror
Privacy-safe
Synthetic twins
Statistically faithful, privacy-safe populations, such as fourteen wealth personas with their micro-personas and signals, used to test journeys and products before launch.
How work flows
Process twins
A model of how work flows, such as subsidy approval or month-end close, used to find the bottleneck and decide where an agent should step in.
In telecom, the operational twin is a living knowledge graph of the network: every element, link, configuration and past fault, updated as telemetry arrives. Agents use it to trace shortest paths, find the most central nodes and cluster alternative routes before customers notice a problem.
02 — Taxonomy in wealth banking
Defined once. Reused by every journey.
Wealth management journey illustration.
The objects behind our ten pre-built advisory journeys, each defined once in Twyn Atlas.
| Object | Classification | What the twin tracks | Journeys it powers |
|---|---|---|---|
| CIO house view | Content › Research › House view › Asset class or theme | Version, stance by asset class, approval status, where it is used, engagement by RMs and clients | 01 House views hub, 03 Client insights, 05 Product insights, 09 Personalised communication |
| Wealth client (synthetic persona) | Person › Customer › Wealth › Persona › Micro-persona | Life events, holdings, risk profile, consent, contact history, likely needs | 02 Client prioritisation, 03 Client insights, 04 Meeting preparation, 06 Recommendations |
| Portfolio | Asset › Portfolio › Advisory or discretionary mandate | Allocation against model, drift, concentration, performance, suitability flags | 07 Portfolio review and rebalancer, 08 RM book insights |
| Structured product | Instrument › Structured note › Autocallable or capital-protected | Barrier levels, coupon and observation dates, maturity, reinvestment window, holders | 10 Structured product lifecycle monitor, 06 Recommendations |
| RM book | Organisation › Relationship manager › Book of clients | Client coverage, contact frequency, pipeline, revenue, follow-ups due | 02 Client prioritisation, 08 RM book insights |
| Campaign | Engagement › Campaign › Segment variant | Audience, content version, channel, opens, replies, meetings and trades that followed | 09 Personalised communication, with monitoring and continuous optimisation |
03 — AI-driven data governance
The foundation every agent inherits.
Each capability is available on its own or as part of TwynTrust.
01
Data quality management
Profiling, AI-suggested rules and scorecards across completeness, validity, consistency, uniqueness, accuracy and timeliness, with remediation agents that open the fix ticket for the right steward.
02
Master data management
Golden records for customers, products, legal entities, charts of accounts, farmers, land parcels and network sites, with ML match and merge — including phonetic matching across Indian scripts.
03
Reference data management
Versioned code lists and crosswalks: currencies, countries, crop and variety codes, tariff plans, cost centres. One approved list, mapped to every system that uses it.
04
Metadata management
Data catalog, business glossary and critical data element definitions, with AI-generated descriptions that stewards approve before publication.
05
Lineage and reconciliation
Source to transformation to store to consumer, traced automatically, with reconciliation agents that match ledgers, stock registers and usage records continuously.
06
Data security and privacy
Sensitive-data discovery and classification, masking and tokenisation, role- and attribute-based access, consent records and alignment to laws such as India's DPDP Act.
07
Data contracts
Formal producer–consumer agreements covering schema, meaning, master-data alignment, quality thresholds, SLAs and versioned change control, enforced before data lands.
08
Natural-language BI
A governed semantic layer so leaders can ask questions in plain language and get answers that use the same KPI definitions as the board pack.
04 — Telecom network operations
Taxonomy in telecom network operations.
The objects behind our network operations agents, each defined once in Twyn Atlas and shared by assurance, planning, field and compliance teams.
| Object | Classification | What the twin tracks | Agents and solutions it powers |
|---|---|---|---|
| Network element | Network asset › Device › Core router, cable modem termination, optical line terminal or customer premises device | Health, uptime, configuration, firmware, latency, packet loss, jitter, signal strength and past faults | Network monitoring, anomaly detection and configuration management agents |
| Service | Product › Service › Residential broadband, Wi-Fi, voice, 5G, enterprise Ethernet or wavelength | The network elements it depends on, SLA, quality of experience and the customers it serves | Service quality and customer impact analysis agents |
| Incident | Event › Incident › Severity › Probable cause | Correlated alarms, affected services and customers, similar past incidents, root-cause confidence | Troubleshooting, root-cause analysis and fault escalation agents |
| Configuration and policy | Policy › Device configuration › Golden template | Intended versus running configuration, drift, change history and policy violations | Configuration management agent; drift and change management |
| Inventory record | Data asset › Network inventory › Planned or live | Cell and site status in planning versus the live network, mismatches and fix history | Twyn Network Inventory Assistant (root cause and self-healing) |
| Field job | Process › Field services › Work order | Remote test results, skills, parts, travel, first-time fix and repeat visits | Twyn Field Assist, dispatch optimisation, predictive maintenance |
| Site and contract | Asset › Site › Lease or rental contract | Landlord, term, renewal date, power and access clauses, amendments | Twyn Telecom Contract Assistant, rental contract management |
05 — Network operations agents
Agentic network operations.
The same governed flow, applied to a network operations centre. Specialist agents share one contextual memory, and an agent supervisor decides who acts next.
| Agent | What it does |
|---|---|
| Data acquisition agent | Collects telemetry over SNMP, gNMI, NetFlow, syslog, TR-069 and REST; identifies each device's protocol, uses a small model to infer schema changes, normalises everything to one structure and updates the network knowledge graph. |
| Network monitoring agent | Tracks latency, packet loss, jitter and link stability against a historical baseline of network health, and watches for drift in policies, networks and devices. |
| Anomaly and incident detection agent | Compares real-time anomalies with historical failure patterns to set severity, then logs, classifies and ranks incidents by urgency. |
| Troubleshooting and root-cause agents | Correlate events across layers, isolate the failing element and attach a confidence score and the evidence used. |
| Performance optimisation and capacity planning agent | Forecasts demand, finds congestion hotspots and proposes rerouting or capacity changes for an engineer to approve. |
| Security threat detection agent | Watches for suspicious traffic and configuration changes, and recommends containment steps. |
| Fault escalation agent | Routes each incident to the right engineering tier with a ready-made evidence pack. |
| Configuration management agent | Compares running configurations with golden templates and drafts corrections and rollbacks. |
| Service quality agent | Tracks quality of service and quality of experience for each service, not just each device. |
| Customer impact analysis agent | Works out which customers and services an incident touches, so care teams can inform them before they call. |
| Agent supervisor | Orchestrates the flow between agents, resolves conflicting recommendations, logs every interaction and keeps the shared memory current. |
Design principles behind the agents:
- Small language models grounded in a network knowledge graph, so agents can run at the edge and in local data centres, not only in the cloud.
- Existing applications become agent tools, including protocol translators, routing calculators, congestion detectors, knowledge-graph mappers, retrieval pipelines and connectors. Nothing that already works is replaced.
- Centralised management, decentralised execution: each agent is a standalone service that scales on its own, and new agents join without disrupting running workflows.
- Retrieval grounded in incident records and device configurations, to raise context and reduce hallucination.
- Cloud, platform and model independent, with open-source components where they cut licence cost.
- Built-in retries, fallbacks and health checks; encryption in transit and at rest; role-based access and input validation on every request.
06 — Our data quality approach
From first principles to quality built in.
Five stages that fix the process, not just the record.
FrameworkAgree the cost of poor data, the critical data elements, dimensions, owners and tooling.
Definition and discoveryProfile the data, define rules and thresholds, and set a measured baseline.
ImprovementCleanse, enrich and complete records, working with the business on every change.
Control and monitoringScorecards, alerts and root-cause analysis that fix the process, not just the record.
AutomationQuality checks embedded in source systems, MDM hubs and data contracts.
07 — Reference architecture
Four layers. Every product.
Technology, cloud, model and prompt agnostic. It runs on the hyperscaler or data platform you already use, routes every model call through a gateway with content safety, and logs every step for lineage and audit.
L4Governance
Role-based access, guardrails against unsafe prompts and jailbreaks, LLMOps and FinOps tracking, audit logs, responsible-AI checks.
L3AI core
Agent orchestration, model gateway, retrieval and grounding, speech-to-text, document intelligence, evaluation.
L2Semantic
Universal taxonomy, ontology and the knowledge base that agents reason over.
L1Data
Ingestion and connectors, data contracts, quality, MDM and RDM, vector and search stores, lineage.
08 — Network operations architecture
Reference architecture for network operations.
How the layers above come together for a telecom operator, with data governance as a layer of its own.
| Layer | What it does |
|---|---|
| Data sources | Customer premises devices, core and edge routers, power and temperature sensors, weather and energy-grid feeds, and the historical data warehouse. |
| Acquisition and integration | Data collection and integration agents, real-time monitoring, protocol translation, a streaming message bus, API integration and a retrieval pipeline. |
| Data foundation | Data lake, data warehouse, network knowledge graph and vector store. |
| AI models and analytics | Demand forecasting, autoencoders for anomaly detection, predictive and deep-learning sequence models, gradient-boosted failure models, reinforcement learning for routing, automated policy-update agents and multi-agent orchestration. |
| Data governance | Data quality, metadata management, master data management, data lifecycle management, and data and regulatory compliance. |
| Agent operations | Monitoring and diagnostics, agent coordinator, model repository, MLOps and SLMOps, vulnerability detection, and template and configuration management. |
| Architectural qualities | Highly scalable, cloud agnostic, edge enabled, decentralised, domain driven and cost efficient. |
Power to act