Recruiting · Production

Upwork scales custom AI models with Uma to cut job post creation time 80% and improve work outcomes

The problem

Upwork needed to serve a large and growing customer base that expected AI solutions, but general-purpose LLMs lacked the precision and contextual awareness needed for marketplace-specific tasks like job post creation and candidate matching.

First attempt

Standard pretrained LLMs produced vague, imprecise responses drawn from broad internet data and were unable to hold conversations that progressively learn about specific user needs, limiting their usefulness for Upwork's marketplace challenges.

Workflow diagram · grounded in source
1
Client triggers job post creation
trigger
“tackle key customer challenges, starting with the difficulty of creating effective job posts for complex projects”
2
GPT-3.5 generates job post
ai_action
“Leveraging GPT-3.5, it streamlines job post creation”
3
Job post published faster
output
“reducing time to posting by 80% for our clients”
4
User query submitted to Uma
trigger
“helping a client determine the next steps in finding a web developer for a pet store business”
5
Uma asks follow-up questions
ai_action
“It takes time to understand what the user is looking for and asks specific follow-up questions to get to the root of the problem so it can best provide an answer and guidance. It also carries a longer conversation, gathering more informa…”
6
Uma retrieves historical signals
ai_action
“Uma is trained on a vast selection of rich historical signals collected on Upwork and therefore can draw from specific examples of how to go about solving a problem”
7
Uma delivers recommendation
output
“it starts by sharing a recommendation to start looking for a web hosting service, knowing this has been a successful strategy to this problem in the past for customers on Upwork”
Reported outcome

The Job Post Generator reduced time to posting by 80% for clients, and the custom Uma model delivers more precise, conversational guidance by drawing from Upwork's historical platform signals, boosting freelancer productivity and work quality.

Reported metrics
Time to posting for job posts80%
Freelancer productivityboost productivity
Freelancer earning potentialincrease their earning potential
Work qualityimprove work quality
Reported stack
UmaJob Post GeneratorUpwork Chat ProGPT-3.5GPT-4ClaudeGPT-4oLlama3.1OpenAI
Source
https://www.upwork.com/blog/scaling-ai-models-for-better-work-outcomes
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The Job Post Generator reduced time to posting by 80% for clients, and the custom Uma model delivers more precise, conversational guidance by drawing from Upwork's historical platform signals, boosting freelancer prod…

What tools did this team use?

Uma, Job Post Generator, Upwork Chat Pro, GPT-3.5, GPT-4, Claude, GPT-4o, Llama3.1, OpenAI.

What results were reported?

Time to posting for job posts: 80%; Freelancer productivity: boost productivity; Freelancer earning potential: increase their earning potential; Work quality: improve work quality (source-reported, not independently verified).

What failed first in this deployment?

Standard pretrained LLMs produced vague, imprecise responses drawn from broad internet data and were unable to hold conversations that progressively learn about specific user needs, limiting their usefulness for Upwor…

How is this recruiting AI workflow structured?

Client triggers job post creation → GPT-3.5 generates job post → Job post published faster → User query submitted to Uma → Uma asks follow-up questions → Uma retrieves historical signals → Uma delivers recommendation.

Related recruiting cases