Tech Mahindra and NVIDIA

AI-Powered Reasoning Comes to Telco NOCs: Tech Mahindra + NVIDIA

Tech Mahindra has partnered with NVIDIA to launch an AI-powered Telco Network Operations Reasoning Agent on the Orion platform. The goal: move Communication Service Providers toward Level 4+ autonomous operations with intelligent, closed-loop execution-without exposing customer data or PII.

Most NOCs still run on rules and manual steps. Engineers burn hours correlating alarms, logs, and KPIs across fragmented systems. This collaboration puts AI reasoning at the core of daily operations so incidents validate, diagnose, and resolve faster-and with more consistency.

What's inside the solution

• AI reasoning agent for autonomous alarm validation, root-cause analysis, and resolution across OSS and BSS.
• Closed-loop execution that can act through existing tools and workflows, with human-in-the-loop controls.
• Extensible design: start with a Large Telco Model, then add domain-specific agents for RAN, transport, core, and IT services.
• Built on NVIDIA AI Enterprise; customized with synthetic and anonymized data via NVIDIA NeMo and deployed as an NVIDIA NIM inference microservice.
• Reported 2-3x accuracy improvement versus a non-fine-tuned model.
• Enterprise-grade privacy: no customer data or PII required.

Why operations leaders should care

• Cut mean time to identify and repair by automating correlation and first actions.
• Reduce alarm noise and ticket churn with reasoning-based deduplication and validation.
• Improve shift-to-shift consistency; fewer escalations, fewer handoff gaps.
• Safer change execution via policy guardrails and rollback logic.
• Better customer experience from faster, predictable incident handling.

How it works at a high level

• Ingest telemetry, alarms, logs, and topology from OSS/BSS and data lakes.
• Normalize context (inventory, services, customers, policies).
• Large Telco Model enriches context; reasoning agent evaluates likely root cause and recommended actions.
• Action orchestration executes playbooks via ITSM, assurance, and network controllers.
• Human-in-the-loop thresholds and approvals for sensitive workflows.
• Feedback loop captures outcomes to retrain and tighten policies.

Practical first moves

• Pick one high-volume incident class (e.g., flapping links, cell degradations, failed changes).
• Codify runbooks as machine-readable playbooks with clear guardrails.
• Integrate with ticketing, change, and notification systems before touching controllers.
• A/B test on historical data; then run in shadow mode; then limited production with rollback.
• Expand domain by domain once auto-resolution accuracy is proven.

Data privacy and governance

The approach uses synthetic and anonymized data to train and refine models. Customer identifiers never enter the loop, helping teams meet privacy and compliance requirements while still gaining AI-driven insight.

Metrics to track

• MTTI, MTTR, and percent of incidents auto-resolved.
• Alarm noise ratio (duplicates/correlated vs. unique).
• Reopen rate and false-positive rate from autonomous actions.
• Change success rate and policy-guarded actions vs. manual overrides.
• SLA breaches per domain and per region.

What to ask your team this week

• Which five incident patterns consume the most engineer time?
• Where does alarm validation stall-data gaps, tooling, or ownership?
• Which runbooks are consistent enough to automate with approvals?
• What telemetry or topology context is missing for reliable RCA?
• What are the hard guardrails we won't cross without human sign-off?

Ecosystem notes

• NVIDIA stack (AI Enterprise, NeMo, NIM) provides the model, tooling, and inference layer.
• Tech Mahindra brings telecom domain expertise, Orion platform integration, and production hardening.
• The architecture is modular-adopt incrementally, scale once KPIs improve.

Get started

If you're planning a phased move to autonomous operations, align your runbooks, guardrails, and KPIs now. For an overview of the enabling platform, see NVIDIA AI Enterprise. To explore the collaboration, visit Tech Mahindra.

Upskilling your NOC and network teams helps close the implementation gap. Start with the AI Learning Path for Network Engineers to build the skills for reasoning agents, closed-loop automation, and data-driven operations.