automation.engineer
All case studies

Services business / Project case study

2023Recorded work

Find relevant opportunities.
Explain the match.
Prepare the opening.

Services business needed a practical way to review potential jobs. A team-built workflow gathered opportunities, explained each relevance decision and prepared a draft opening, with editable search criteria and experience examples for human review.

Client
Services business
Contribution
Team contribution, criteria and client setup guidance
Project status
Team-built workflow & handover

Tools & working context

Project scope and implementation.

  • Make
  • historical RSS feeds
  • GPT-4
  • Airtable
  • prompt examples
  • human review

01 / Business context

The problem
to solve.

A list of job posts still leaves a reviewer with several tasks: decide whether each opportunity fits, understand why it was selected and find a relevant starting point for an application.

Historical job searches and prompts containing examples handled the first review pass. People still chose the opportunities and submitted the applications.

Inside the workflow

A closer look
at the handoffs.

Start with the main path and its branches. Select a step for context, or switch to the technical map for implementation detail.

  • Historical service search feeds
    Multiple defined RSS searches supply matching jobs for the service areas.
  • Editable examples and services
    Positive and negative examples supply relevance context; experience informs drafts.
  • Assess relevance with a reason
    Use example-based GPT-4 prompts to explain why a post does or does not fit.
  • Prepare personalized opening
    Generate an application starting point using service and experience context.
  • Airtable opportunity records
    Keep the post, reason, draft and review state in the team’s shared queue.
  • Team reviews the opportunity
    A person decides whether the post merits an application.
  • Human application
    The team chooses and sends applications after review.
  • Record review outcome
    Keep the reason a post was not applied to so the selection can be examined.
  • Adjust the selection examples
    Use review outcomes to refine editable prompts and search criteria.
Read the workflow notes
  1. 01 / Collect

    Retrieve the defined searches

    Historical RSS searches supplied job posts matching the configured service areas. The client adaptation focused those searches on the services the business actually described.

  2. 02 / Explain

    Classify fit and draft an opening

    GPT-4 prompts used examples to produce a relevance result and reason, followed by a personalized opening. Structured outputs supplied fields for the next modules.

  3. 03 / Review

    Keep criteria and decisions editable

    Airtable stored the posts and drafts. The handover explained experience examples, criteria changes and a feedback loop for refining classification.

This is a historical integration account; availability of the original RSS interface today is not assumed. Applications were reviewed and submitted by people.

03 / Implementation detail

How it
was built.

Each relevance result included a reason. Reviewers could spot unsuitable selections and use those examples to adjust the matching instructions.

Nick and other contributors built parts of the system. Noah contributed selection criteria, guidance and adaptations within that team effort.

Moving the setup into the client’s accounts was still pending in the handover. Instructions for sharing access did not confirm that the move had happened.

04 / Outcome & scope

Result
and scope.

The team-built workflow prepared job matches, relevance explanations and draft openings for people to review, with guidance for client setup.

No automated submission, fine-tuned model, classification-accuracy metric, job wins or conversion improvement is claimed.