What challenges did Nexthink face before AI implementation?

Make complex DEX analytics and NQL accessible to non-technical users through natural language. Increase query accuracy and consistency while scaling labeled data creation for model training. Reduce token usage, inference costs, and manual annotation burden.

Source

Evidence: Low35/100

Nexthink builds enterprise AI agents using fine-tuned LLMs with Amazon SageMaker AI

Use case typeLegal AI assistantUpdated Jun 13, 2026

Nexthink, a Swiss digital employee experience platform, built Nexthink Assist to let users ask natural-language questions and receive DEX insights without needing to know Nexthink Query Language (NQL). The team first used prompt engineering and RAG, then fine-tuned a specialized 7B LLaMA model on a golden dataset to improve accuracy, consistency, and coverage across the data model. They also built an automated labeling workflow using Amazon SageMaker Ground Truth and Amazon Bedrock with Claude 3.5 Sonnet as an LLM-as-a-judge, supported by S3, AWS Lambda, and Amazon SQS.

Organization: Nexthink
Industry: Tech & Comms
Location: Switzerland
Published: February 2026

Reported outcomes

Accuracy: Approximately 30% higher

Quality & accuracy

Catalog median for quality & accuracy deployments: +40% across 55 reported metrics. Compare benchmarks →

Why do we believe this?Outcome claims, sources, and evidence checks

Normalized claim

Accuracy: 30% increase

AWS BlogFeb 11, 2026Blog postInferred claimLow evidence strength

Improved query accuracy by about 30%.

Normalized claim

Cost: 80% decrease

AWS BlogFeb 11, 2026Blog postInferred claimLow evidence strength

Reduced token usage and inference costs by about 80%.

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Nexthink
Provider: AWS
Maturity: Unknown
Linked source: AWS Blog

No explicit deployment-stage evidence found.

Customer identity supportedSource describes one deploymentMaturity evidence evaluated

Primary read

Use case focus

Showing 3 of 6

1Generative AI for enterprise search
2AI agent
3Natural language to query translation

Built NL2NQL to translate natural-language questions into NQL queries.
Iterated from prompt engineering to Retrieval Augmented Generation and then fine-tuned a specialized 7B LLaMA model using Amazon SageMaker AI with LoRA and distributed training.
Used Amazon SageMaker Ground Truth for SSO-based annotation workflows and Amazon Bedrock with Claude 3.5 Sonnet as an LLM-as-a-judge to automate labeling quality checks.
Orchestrated data preparation and labeling flow with Amazon S3, AWS Lambda, and Amazon SQS.
Used SageMaker model deployment and monitoring capabilities for the fine-tuned model.

Technologies

Amazon SageMaker AI Amazon SageMaker Ground Truth Amazon Bedrock Amazon SageMaker JumpStart Llama Low-Rank Adaptation (LoRA)Amazon S3 AWS Lambda Amazon SQS

Improved query accuracy by about 30%.

Architecture

An S3-based annotation pipeline triggers AWS Lambda preprocessing, routes tasks through Amazon SQS to Amazon SageMaker Ground Truth for SSO-authenticated labeling, stores standardized labels back in S3, and uses Amazon Bedrock Claude 3.5 Sonnet for automated labeling/LLM-as-a-judge before fine-tuning a 7B LLaMA model in Amazon SageMaker AI with distributed training and LoRA.

Sources & evidence1

Evidence: Low35/100Evidence strength