Autonomous networks are quickly becoming certainly one of the highest priorities in telecommunications. In response to the newest NVIDIA State of AI in Telecommunications report, 65% of operators said AI is driving network automation, and 50% named autonomous networks as the highest AI use case for ROI.Â
Yet many telcos still report gaps in AI and data science expertise. This makes it difficult to scale secure, closed-loop automation across complex, multidomain networks. Â
Most telecom network operations centers (NOCs) today operate using reactive, alarm-driven workflows. Engineers manually triage 1000’s of incidents across multiple tools, sift through a high volume of alarm and performance data, and stitch together fragmented dashboards and logs before applying a fix or dispatching a field team. NOCs are a natural start line for autonomous networks, because they concentrate high-volume, repeatable tasks where AI can directly cut MTTR and OPEX.
Tech Mahindra, a number one global provider of technology consulting and digital solutions to enterprises across industries, and NVIDIA are collaborating to shut this AI skills gap. They’re doing so by making autonomous network constructing blocks—open models, tools, and implementation guides—into assets telecom developers can readily adopt and adapt in their very own environments.Â
This post outlines methods to positive‑tune reasoning models with NVIDIA NeMo in order that they behave like NOC engineers, safely driving closed‑loop, self‑healing workflows. It shows methods to:Â
- Generate synthetic, telecom‑realistic incident data
- Translate expert procedures into structured reasoning traces using the production-grade reference workflows. This teaches the model to coordinate tools, reason over network state, and execute fault‑management tasks end to finish
The result’s a repeatable method that telco teams can use to construct their very own specialized AI agents for network operations. These agents can perform triage, root‑cause evaluation, and determination for prime‑volume incident classes, helping operators progress toward TM Forum Level 4 highly autonomous networks and beyond.
Why do network operations centers need reasoning models?
Traditional NOC automation is generally rule‑based and open‑loop: scripts trigger on fixed conditions but struggle with noisy signals, cross‑domain dependencies, and always changing network behavior. Because of this, many Level 1 and Level 2 tasks—triage, root‑cause evaluation, validation after a change—still depend upon manual effort, keeping MTTR high and limiting how far operators can move toward truly autonomous operations.
A telco reasoning model becomes the engine for an AI agent that may tackle this work pattern in a controlled, auditable way. As a substitute of hard‑coded runbooks and point scripts, the agent uses the model to interpret incidents, resolve which tools to call, and adapt its actions based on live responses. Key features include:
- AI reasoning plus tool-calling: Replaces manual alarm triage by invoking NOC tools for validation, root‑cause evaluation, and remediation across existing systems
- End-to-end automation: Handles alarm validation, RCA, and healing for various incident types akin to outages, flaps, congestion, and configuration issues
- Noise reduction: Filters self‑clearing or low‑value alarms using historical patterns so engineers can concentrate on higher priorities
- Resolution in seconds, not hours: Shrinks resolution time for prime‑volume, well‑understood incidents from hours to seconds, significantly reducing MTTR
The consequence is a closed‑loop, self‑healing network. Specialized NOC agents handle routine triage and determination, and engineers shift from reactive alarm handling to proactive optimization and sophisticated problem-solving.
Designing a telco reasoning pipeline
The technical approach to this solution combines the next components into one reproducible pipeline:Â
- Synthetic incident data
- Expert NOC procedures
- Structured reasoning traces
- Supervised positive‑tuningÂ
- EvaluationÂ
As a substitute of attempting to learn from raw logs and alarms directly, the model is trained on curated examples that show how an experienced engineer would analyze an incident, call tools, and choose when a fix is complete.
On this case, Qwen3-32B is the bottom reasoning modeling that’s fine-tuned for telco NOC workflows using the next design principles:
- Specializing in a small variety of high‑impact faults, which account for nearly all of incidents and require deliberate motion. This permits the model to learn deeply on the fault classes that matter most.
- Defining step-by-step operational guidelines for every problem type including RCA and remediation steps and NOC tools that agents must use.
- Generate synthetic reasoning traces that capture multistep tool calls and the rationale behind each decision, using the NeMo Skills reference workflow to automate trace and incident generation.
NeMo Skills orchestrates this pipeline end to finish, using its CLI, vLLM or TensorRT LLM servers, and training utilities to maneuver from raw incidents to a fine-tuned telco reasoning model.​
The input to the pipeline is a totally synthetic incident dataset that’s modeled on real NOC behavior. Each record includes fields akin to region, domain, priority, problem type, possible cause, and time stamps. Engineer notes are also included, describing intermediate steps and shut notes summarizing the ultimate resolution and shut code.Â
An incident summary captures why the network was degraded or down and is the backbone of what the model is trained to unravel. The pipeline concentrates on probably the most frequent, high-impact faults that account for the majority of incident volume and require explicit motion. The reasoning model learns deeply on the cases that drive MTTR and OPEX.
To model realistic NOC workflows, a set of custom tools are defined for agents to call in multistep procedures, akin to:
- Acknowledging and tracking the initial alert
- Checking site and equipment status
- Performing distant actions (reset, unlock, enable)
- Monitoring for automatic recovery or alarm clearance
- Checking topology, power, and fiber, plus public outage information
- Applying configuration fixes
- Rechecking alarm status when it stays energetic
- Investigating persistent or recurring alarms
- Documenting actions and standing updates
- Coordinating onsite dispatch or hardware substitute
- Confirming final site health and shutting the incident
For every problem type, domain experts translate existing workflows into step‑by‑step guidelines that map onto these tools. Examples include which triage toolkit to seek the advice of first; which alarms to question; when to reboot a tool; and methods to confirm a fiber cut, power outage, or network element faults.Â
These guidelines turn out to be blueprints for the synthetic reasoning traces the model will learn from. They later define the motion space that NOC agents use when executing closed‑loop workflows in production.Â
Turn expert procedures into reasoning traces
To show expert NOC procedures into training data for a telco‑specialized reasoning model, follow the three-step NeMo Skills workflow outlined below. It converts runbooks into structured, multiturn reasoning traces ready for autonomous NOC agents.
Step 1: Generate structured motion sequences
Using a reference workflow from NeMo Skills, a teacher model generates standardized motion sequences for every incident based on prompts that include incident fields and guideline templates. The steps map on to NOC tools.
Traces are formatted so each step records the motion, its parameters, the tool call, and the immediate result, forming a structured view of the NOC workflow.​
Step 2: Attach per‑step reasoning
A second pass enriches every motion with reasoning text that explains why the step is taken, what signals it uses, and the way it influences the following decision. This creates a series of reasoning that reflects how an experienced NOC engineer reasons over topologies, alarms, and historical behavior.Â
Because raw traces could be verbose or repetitive, a squashing phase merges related steps while preserving key decision points, making sequences more efficient for training.
Step 3: Formatting for multiturn, tool‑calling models
Using one other workflow from NeMo Skills, the formatted traces are converted right into a Qwen-compatible format that encodes each the dialogue-style interaction and tool-calling actions over multiple turns. Multiturn tokenization simulates realistic interactions where the agent alternates between reasoning, calling tools, and interpreting tool responses, which is crucial for deploying a ReAct-style NOC agent.​​
The result’s a curriculum-structured dataset where easier cases and shorter traces appear earlier, while more complex multi-step incidents appear later, supporting curriculum learning during model training.​​
Tremendous-tuning the telco reasoning modelÂ
The fine-tuning phase uses an ordinary train/test split on the compiled reasoning dataset, with NeMo Skills orchestrating data preparation and Qwen3 32B serving as the bottom reasoning model. NeMo Skills prepare_data utilities apply a telco‑specific prompt template (noc_reasoning_sft) and the Qwen tokenizer. This makes each trace within the training split right into a supervised positive‑tuning (SFT) example that features:
- Incident context and NOC signals
- Multistep tool calls and intermediate results
- Reasoning traces explaining each decision
- Final resolution and incident summary
This produces a single JSONL file of SFT-ready examples for the telco reasoning model.​
To enhance learning efficiency, curriculum learning is applied by ordering samples from easy, single‑problem incidents to more complex multistep, multitool cases. This permits the model to master core NOC behaviors before tackling long, multiturn troubleshooting patterns.Â
Multiturn tokenization ensures that every example preserves realistic sequences of queries, tool calls, responses, and follow‑up actions, slightly than isolated single‑turn prompts. These capabilities are critical for downstream ReAct‑style agents that must coordinate multiple tools over long contexts.
Ultimately, Qwen3‑32B is positive‑tuned on this telco reasoning curriculum with long sequence lengths and tensor model parallelism across GPUs. Checkpointing and experiment tracking allow teams to iterate on data quality, curriculum design, and hyperparameters.Â
The result’s a telco‑specialized reasoning model that understands incident fields, close codes, and NOC procedures, and may reliably drive multitool, multiturn tool‑calling workflows in production.
Evaluating incident summary accuracy and safety
Initial evaluation focuses on incident summary accuracy: how well the model, embedded in a ReAct‑style agent with tools, predicts and executes the right resolution path for a given incident.Â
Experiments compare the positive‑tuned telco reasoning model against a baseline Qwen3‑32B on held‑out incidents, measuring accuracy, precision, and recall across problem and shut‑code categories. Incident summary accuracy may also be analyzed inside a single problem type to focus on where reasoning traces and curriculum learning deliver the most important gains, informing future iterations of synthetic data generation and guideline design.
Evaluations across multiple iterations show that the fine-tuned model improves accuracy from roughly 20% to 60%.
Beyond incident summary metrics, additional evaluation methods could be introduced over time to further harden the system, including:
- LLM‑as‑a‑judge setups to judge reasoning traces for correctness, completeness, and safety
- LLM‑as‑a‑judge to evaluate final conclusions and remediation plans
- Tool‑calling benchmarks akin to BFCLv3 to measure how reliably the agent sequences and interprets tool calls
- Rollout and rejection sampling to emphasize‑test behavior across many simulated incidents
- Controlled errors injected into traces to show the model to detect and recuperate from its own mistakes
- Incorporation of retrieval‑augmented generation (RAG) with historical few‑shot examples to enhance robustness on long‑tail scenarios
Start constructing telco reasoning models for autonomous networks
Telco‑specific reasoning models—powered by synthetic data, structured traces, and secure tool‑calling—can move NOCs toward zero‑touch, self‑healing operations. By specializing in high‑impact close codes, encoding expert guidelines as multiturn reasoning traces, and positive‑tuning large models with the NVIDIA NeMo software toolkit, operators can construct agents that reliably tackle real NOC engineer tasks.Â
The pipeline is reusable and adaptable, so this approach could be tailored to every operator’s tools, data, and policies. This accelerates the industry’s transition from manual alarm handling to intelligent, autonomous network operations.
To start fine-tuning a reasoning model to construct AI agents for network operations, see Teaching a Model to Reason over Telecom Network Incidents.
