Customer Stories

Building Resume Writing AI Agents for Talent Inc.

Career services
Foundation model
resume writing
2 resume writing AI Agents

Two premium-quality AI Agents

4.67x faster

Time to complete one resume decreased from 3.5 hours to 45 minutes


The percentage of Talent’s workflow that’s now automated with AI Agents

Talent Inc.
“I could not be happier with where we are right now. It's by far a net positive for us, and I don't see that journey really slowing down.”
Byron Matthews
CEO @ Talent Inc.


  • Manual resume writing is slow, resulting in a fairly high cost per order and potential customer frustrations;
  • Automating it is risky, as it could lead to a decrease in quality and customer satisfaction


  • Customization of GPT-3.5 and GPT-4
  • In-context learning and fine-tuning with Talent’s data
  • 2 endpoints for each AI Agent: resume writer and feedback


  • 2 resume writing AI Agents
  • Reduced time to complete one resume (3.5 hours → 45 minutes)
  • Increased conversion rates
  • Improved resume quality


Talent Inc. is a career services company specializing in writing professional resumes for job applicants. 

  1. They have an experienced, skilled team of writers, but the workflow is slow. On average, manually completing just one resume takes 3.5 hours—or 43.75% of the entire 8-hour workday.
  1. Talent also provides a DIY builder that helps customers generate polished resumes themselves. However, they realized they could offer users more value by helping them generate resume content instead of just transferring it onto a nice template.

In May 2023, they partnered with us to build two AI Agents that could help them reach those goals. We delivered the first one in July and the second one in October.

Today, Talent generates resumes 4.67x faster than before, converts more customers, and is working with us on developing another set of AI Agents for other services. 

The Dream: Improving Efficiency Without Decreasing Quality

As a career service company, Talent relies on many manual processes to get work done. Most notably, their resume-writing service was mostly manual and slow. They realized that Generative AI could help them accelerate it while also potentially increasing customer and employee satisfaction. 

More specifically, they started considering automating two of their services: and TopResume. is a self-serve/DIY resume builder that customers can use to quickly transfer the content of their resumes onto a professional-looking template. 

An example of how populates resume templates with users’ inputs. 

However, it offered limited added value to users—users still needed to input all information and make sure it was well-written. did not improve nor generate new content for them. This also made the process time-consuming, requiring a lot of effort from customers.

Problem #1

Limited added value

Problem #2

Time-consuming and laborious process for users

TopResume, on the other hand, is a premium resume-writing service provided by expert human writers. It promises higher-quality resumes but is difficult to scale. Talent’s writers need about 3.5 hours to finish a single resume—or an entire, 8-hour workday to deliver roughly two resumes. 

Besides being laborious and costly, the process also led to longer-than-ideal wait times and, potentially, frustrated customers.


Problem #1

Time-consuming and laborious for employees = limited scalability and a high cost per order

Problem #2

Long wait times and decreased customer satisfaction

By upgrading these two services with AI, Talent was hoping to decrease cost per order, increase writer efficiency, and improve customer satisfaction. However, in order to hit those goals, it was crucial that the quality of resumes remains the same. 

The main challenge was maintaining the same degree of quality while minimizing human intervention. 

Talent needed a reliable and highly accurate AI to overcome this challenge. Despite already having an existing data science team, they partnered with us to finish the project faster and scale their initiatives more easily. 

High-Quality LLMs + High-Quality Training

We needed to take two main steps to ensure that the resume quality remains the same or, preferably, increases with automation: 

  • Step 1: Choose suitable, high-quality foundation LLMs, especially for TopResume.
  • Step 2: Use Talent’s data to align the model with their use case and writers’ established processes.

Once we were happy with the outputs, there was only one last step to take:

  • Step 3: Integrate the newly developed AI Agents with Talent’s products.

Here’s a brief overview of how we approached each step.


We chose GPT-3.5 and GPT-4 as foundation models for and TopResume respectively. 

  • GPT-3.5 is significantly cheaper and faster and provides a satisfactory, but not premium level of quality.
  • GPT-4 is more advanced and generally provides higher-quality outputs. It also has a bigger context window, allowing for more detailed inputs.
GPT-3.5 GPT-4
Accuracy Medium (e.g., 70.0% on the MMLU benchmark) High (e.g., 86.4% on the MMLU benchmark)
Speed Higher (according to some estimates, maximum flow rate is 108.94 tokens/second) Lower (estimates indicate a maximum flow rate of 12.5 tokens/second; approx. 8.71x lower than GPT-3.5)
Cost Lower (ranging from $0.0010 to $0.0020 per 1K tokens) Higher (ranging from $0.03 to $0.12 per 1K tokens)
Context Window Lower (~10,000 words of context) Higher (~25,000 words of context)

It was clear that GPT-4 was better-suited for Talent’s premium service (TopResume), while GPT-3.5 proved to be sufficient for their DIY service (


After selecting the models, we proceeded to conduct in-context learning using data from both Talent and public sources.

Training data

  • Input-output pairs (Talent’s questions on customers’ backgrounds and aspirations + customer-submitted answers)
  • Additional data (e.g., from LinkedIn)


  • Input-output pairs (user-submitted information + corresponding resumes written by Talent’s staff)
  • Human-written critiques of existing resumes
  • Additional data (e.g., from LinkedIn)

In-context learning helped us align the models with Talent’s use case and ensure they provide satisfactory outputs—ones that are almost identical to resumes written by seasoned human writers.

In-context learning

→ allows models to solve novel tasks by providing examples of previous successful task completions

Overview of the process:

1. Provide batch of relevant input-output examples to the model

2. Define the context (e.g., explain the goal of the resume and what it should look like)

3. Repeat 

4. Provide the model with the input data and ask for the output

5. Evaluate performance

6. Repeat the process if needed

We are currently also fine-tuning the models to further improve their performance.


After extensive performance monitoring and evaluation, we integrated the AI Agents with Talent’s products. 

The AI Agent powered by GPT-4 was integrated with TopResume due to higher-quality outputs. It currently generates the first version of resumes that Talent’s writers review and edit before sending them back to customers.

The medium-quality AI Agent, powered by GPT-3.5, was integrated with End users can use it if they select the option to generate an AI-assisted draft, but they can also use the service in the same way they did previously—with no AI involved.  

We’ve created two endpoints for each service: the resume writer endpoint and the feedback endpoint.

  • Resume writer endpoint receives client data as inputs and returns resumes as outputs.
  • Feedback endpoint receives the final versions of resumes edited by Talent’s staff.

That way, the Agents don’t just generate resumes but also learn and improve from edited resumes, providing increasingly higher-quality outputs over time.

Time Savings, Increased Conversions, and Improved Quality’s new feature—called “AI Assisted Draft”—was rolled out to end users in October 2023. However, Talent’s Chief Product Officer, Christian Dwyer, says that ⅓ of their users are already selecting this option rather than starting with a blank template or uploading existing resumes. 

He also mentions that users are now completing their resumes more quickly and converting at a higher rate than users from alternative entry points. 

TopResume’s AI Agent, on the other hand, is already saving Talent’s writers a significant amount of time: roughly 78.57%. It has also successfully overcome the main challenge—preventing a decrease in quality. According to Christian, the quality of resumes has actually increased since they’ve implemented AI into their workflow.

“Being able to go from three and a half hours to 45 minutes while the quality goes up is unbelievable. I think that’s because we use technology to augment human skills.” - Byron Matthews, CEO @ Talent Inc.

As Talent’s CEO mentions, these results can largely be attributed to aligning the AI Agents with Talent’s existing workflows and processes. By ensuring easy adoption, we’ve also given the employees the best chances of using the AI Agents to their advantage. 

Current results are more than promising, even as we continue to monitor and further improve the performance of these AI Agents. 

We are also in the final stages of developing three new, specialized Agents for Talent, which will help forge a new path for the entire industry. 

However, even if competitors start taking the same route, Talent will remain a pioneer in its field, making it difficult for others to outdo their AI initiatives at a later point. Now, it all comes down to further evolving the AI Agents, with the numbers clearly showing we’re already on the right track.

Career services
Foundation Model
resume writing

Achieve enterprise-wide workflow automation

Automate workflows?

Schedule a free,
30-minute call

Explore how our AI Agents can help you unlock enterprise-wide automation.

See how AI Agents work in real time

Learn how to apply them to your business

Discuss pricing & project roadmap

Get answers to all your questions