It support · Production

Linde builds internal employee chatbot with NLP training data from super.AI

The problem

Linde's voice-based chatbot could not reliably recognize user intent or named entities in natural language queries, causing employees to receive irrelevant information rather than useful answers from the internal knowledge base.

First attempt

The existing voice chatbot failed to understand user intent and named entities, rendering it unable to serve relevant information to employees.

Workflow diagram · grounded in source
1
Linde submits training inputs
trigger
“Linde provided a list of entities, utterances, and ground truth”
2
AI classifies utterances
ai_action
“Using Super.Classify, super.AI's no-code solution for data classification, Linde was able to quickly create an accurate training dataset”
3
Text and query augmentation
ai_action
“the data program produces many similar queries with the same intent but different wording”
4
Human verification of variations
human_review
“Involving human workers to both generate phrase variations and verify them against the original input for intent similarity”
5
Training data delivered
output
“Produced accurate training data in 6 weeks from kickoff to final delivery”
6
Chatbot responds to spoken queries
output
“Linde now has an internal knowledge base that quickly responds to spoken user queries posed in natural language”
Reported outcome

super.AI produced accurate NLP and NER training data in 6 weeks, generating 88,000+ phrase variations and enabling 6M+ automations; Linde successfully launched the internal chatbot.

Reported metrics
Training data delivery time6 weeks
Throughput and speed satisfactionvery happy with the throughput and speed
Reported stack
Super.ClassifyNLPNER
Source
https://super.ai/case-studies/world-leader-in-hydrogen-production-builds-chatbot-solution-with-super-ai
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

super.AI produced accurate NLP and NER training data in 6 weeks, generating 88,000+ phrase variations and enabling 6M+ automations; Linde successfully launched the internal chatbot.

What tools did this team use?

Super.Classify, NLP, NER.

What results were reported?

Training data delivery time: 6 weeks; Throughput and speed satisfaction: very happy with the throughput and speed (source-reported, not independently verified).

What failed first in this deployment?

The existing voice chatbot failed to understand user intent and named entities, rendering it unable to serve relevant information to employees.

How is this it support AI workflow structured?

Linde submits training inputs → AI classifies utterances → Text and query augmentation → Human verification of variations → Training data delivered → Chatbot responds to spoken queries.

Related it support cases