Quality assurance · Production

LinkedIn automates search typeahead quality evaluation with GenAI using OpenAI GPT on Azure

The problem

LinkedIn's typeahead search quality assessment relied on human evaluation that was difficult to sustain at scale given platform growth, with manual evaluations involving multiple human evaluators taking days or weeks to complete.

Workflow diagram · grounded in source
1
New experiment triggers evaluation
trigger
“For a new experiment on typeahead relevance, we perform GenAI quality evaluation via the following steps”
2
Typeahead responses collected
integration
“Generate requests with the experiment configs on the Golden Test Set, then call Typeahead backend”
3
GPT prompts generated
integration
“Generate Prompts for GPT 3.5 Turbo on the responses suggestions”
4
GPT batch quality scoring
ai_action
“For each suggestion in a typeahead session, GPT scores it as either 1 (high) or 0 (low)”
5
Quality scores calculated
output
“Post processing GPT responses to calculate TypeaheadQuality scores”
6
Continuous quality benchmarking
feedback_loop
“This metric allows us to monitor and benchmark the health of the typeahead experience over time, helping us quickly identify areas for improvement and measure the impact of new experiments”
Reported outcome

The GenAI Typeahead Quality Evaluator reduced evaluation time from days or weeks to a few hours, and a representative experiment demonstrated a 6.8% absolute improvement in the typeahead quality score at position 10, corresponding to a 20% reduction in low-quality suggestions.

Reported metrics
TyahQuality@10 absolute improvement6.8%
Low-quality suggestions reduction20%
Evaluation turnaround time (new system)just a few hours
Evaluation turnaround time (manual baseline)days or even weeks
Reported stack
OpenAI GPTGPT 3.5 TurboAzure
Source
https://www.linkedin.com/blog/engineering/ai/automated-genai-driven-search-quality-evaluation
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The GenAI Typeahead Quality Evaluator reduced evaluation time from days or weeks to a few hours, and a representative experiment demonstrated a 6.8% absolute improvement in the typeahead quality score at position 10,…

What tools did this team use?

OpenAI GPT, GPT 3.5 Turbo, Azure.

What results were reported?

TyahQuality@10 absolute improvement: 6.8%; Low-quality suggestions reduction: 20%; Evaluation turnaround time (new system): just a few hours; Evaluation turnaround time (manual baseline): days or even weeks (source-reported, not independently verified).

How is this quality assurance AI workflow structured?

New experiment triggers evaluation → Typeahead responses collected → GPT prompts generated → GPT batch quality scoring → Quality scores calculated → Continuous quality benchmarking.