A keyword match and a reasoned fit are different claims
Assigning students to opportunities based on keywords produces long lists that students ignore. A student who has never written the word on their profile might be the best candidate. A student whose profile matches all the terms may not be suitable for the job.
To make a reasonable assessment, you look at the opportunity and at the evidence the student has provided that they have done something similar, for example a project, a module outcome, or a previous placement. You then state that evidence so the student can include it in an application.
Fits are mutual. A match can still be unsuitable even if a person has the necessary skills, because it involves travel or specific hours. If a match fails to take these constraints into account, it wastes the time of both the student and the employer.
The agent puts forward suggestions, with written reasons, to a careers adviser. The agent does not rank any student against another, does not contact any employer, and does not submit any application.
How the agent is built in Agent Studio
Placement policy is determined by three factors: the requirements of your program, the demands of employers in your network, and any limitations a student may specify, which must be treated as absolute.
The system prompt prohibits any inferences about protected characteristics and prevents the use of demographic data when matching. Reasoning is based only on stated skills, previous work, and declared preferences.
Each match must name its evidence and its source, and state what is missing, so the student can see what needs strengthening rather than only what they lack.
The system prompt also requires the agent to indicate when no items match. A blank result is more useful to an adviser than one that has been stretched.
Student profiles are retrieved from Airtable, with access restricted to students who have opted in to matching. Opportunities are received through an HTTP Request node from your partner feeds.
The Agent Chat node produces the reasoning. Each student has one section in Google Docs to hold the adviser's view.
All student-facing matches are gated by a Human in the Loop node, so no match can reach a student or an employer unless an adviser releases it.
What this agent is built from
- Prompt skills: program placement policy, partner requirements, and absolute student constraints.
- System prompt: no demographic data and no inferences about protected characteristics.
- System prompt: each match names its evidence and what is missing.
- System prompt: an empty result is reported rather than stretched.
- HTTP Request (Util) with Airtable (Integration): opportunity feeds and opted-in student profiles.
- Google Docs (Integration): the adviser's view, one section per student.
- When an ActionFlow is enough: If filtering opportunities against a student's declared field and constraints is the work, that workflow is a fair first pass.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.