Build a Claude-Powered Candidate Screener for HR Teams

Give yourself more time for the human side of hiring

When applications arrive faster than you can review them, strong candidates can get lost in the backlog. In hospital HR, that might mean overlooking a nurse with ICU experience, the right certification, and a willingness to relocate.

Claude can help you turn your screening criteria into a practical first-pass tool. You can ask it to build a browser-based application that organizes candidate information, calculates transparent match scores, and tracks your review decisions. It should support your judgment—not choose whom to hire.

A worked example, ClinIQ Screener, reported a build time of 10 minutes and 40 seconds using a prepared prompt. That illustrates how quickly you can prototype; it does not mean your own tool will be production-ready in ten minutes.

Here’s how to define the workflow, prompt Claude, and test the result before you consider using real candidate data.

Start with your screening process—not the code

You don’t need to write HTML or JavaScript to get started. Your expertise is knowing what your hiring workflow needs.

The ClinIQ Screener example asked Claude to build five capabilities:

1. Configure the role: Set the job title, department, experience, skills, cities, languages, and certifications.
2. Add candidates: Upload one CV, multiple CVs, or an existing Excel candidate database.
3. Score against criteria: Calculate a match score using configured requirements.
4. Track reviews: Mark candidate records as shortlisted, rejected, or on hold.
5. Export or reset: Download results to Excel or clear records for a fresh search.

That gives Claude a concrete specification instead of a vague request to “make a recruitment app.”

Before you write your prompt, define what each criterion means. Does “six years’ experience” mean total employment or relevant nursing experience? Is a certification essential, or can a candidate complete it after joining? Is location mandatory, or would relocation work?

Clear criteria make your prototype easier to test and help you screen more consistently.

Give Claude a full workflow to build

Use this prompt as a starting point. It includes privacy and review safeguards that are useful to add to your own prototype.

Build a browser-based recruitment screening prototype for my HR team called ClinIQ Screener.

Include five sections:

1. Role configuration: Role title, department, minimum relevant experience, required skills, acceptable locations, job-relevant languages, and certifications. Let me configure criterion weights and distinguish mandatory requirements from preferences.
2. Candidate input: Single CV upload, multiple CV uploads, and Excel upload. Support PDF and Word files where feasible, and clearly explain unsupported formats or dependencies.
3. Screening: Extract candidate details and calculate a transparent match score. Show the evidence used for each criterion. Mark missing or uncertain information as “needs verification,” rather than assuming the candidate lacks it.
4. Results: Display a sortable table with candidate name, role, experience, location, languages, certifications, score, and review status. Let me manually mark candidates as shortlisted, on hold, or rejected. Do not make final hiring decisions automatically.
5. Export and reset: Export results and review statuses to Excel. Ask for confirmation before clearing data.

Generate the application code and explain how to run it. State parsing limitations, whether candidate data leaves the browser, any external libraries required, and whether offline use is supported. Include sample data for testing.

Start with synthetic or anonymized profiles. Before testing with real applications, check your organization’s privacy, security, and AI-use requirements.

Test every input mode before relying on results

A polished interface doesn’t prove that the screening logic works. Start with a small dataset you can inspect manually, then test each way candidates can enter the tool.

Test one CV for extraction and scoring

One example used a nursing role requiring two years of experience, Mumbai as the location, Marathi and Hindi, and BLS training—Basic Life Support.

The uploaded candidate received a 65 out of 100 match score. The tool identified 17 years of experience and BLS training, but didn’t extract location or language information.

That result makes an important point: a missing detail in a CV isn’t necessarily a missing qualification. Check the original document and, where appropriate, ask the candidate to confirm. A match score reflects alignment with your configured criteria; it is not a probability of job success.

Test multiple CVs for sorting and role mismatches

A second test processed 18 CVs against a nursing search configured for six years of experience, locations including Mumbai, Bengaluru, and Hyderabad, and ACLS certification—Advanced Cardiovascular Life Support.

The tool displayed results in score order. The upload also included a general duty assistant (GDA) profile by mistake. That candidate appeared in the results, giving the reviewer a chance to spot the role mismatch.

Test this on purpose. Does the tool flag a wrong-role profile clearly? Can you inspect why it received its score? Do status counts update when you shortlist someone or place them on hold?

Test Excel filtering against records you know

The Excel test used a database of 30 candidates: 20 nurses and 10 GDAs. With GDA selected as the role, the tool returned the 10 GDA records and excluded the nurses.

That demonstrates role filtering in this particular test. It doesn’t show that all ten candidates met every hiring requirement.

Before using a larger spreadsheet, check column mapping, blank cells, duplicate records, and inconsistent job titles.

Fix problems with specific follow-up prompts

Claude’s first version may contain parsing errors, broken controls, or unexpected scoring behavior. Describe the issue and the change you need.

The candidate’s certification appears in the CV but is missing from the extracted record. Review the parser, explain the likely cause, and update the tool. Add a manual correction option so I can fix extracted fields before scoring.

The example’s CV parser used heuristic rules, including regular expressions. Those methods won’t reliably understand every layout, scanned document, or unusual wording.

The generated tool was described as downloadable and usable offline. Verify that yourself. PDF, Word, and Excel handling may depend on external libraries, so a browser-based application isn’t automatically self-contained or private.

Measure whether the prototype helps your HR work

The reported build time describes a prototype result. It doesn’t establish screening accuracy, fairness, or readiness to replace an applicant tracking system.

For a pilot, track:

• Time spent reviewing a fixed batch of applications.
• Accuracy of extracted fields.
• Qualified candidates missed or incorrectly ranked.
• How often you need to correct extracted information.
• Whether the scoring explanations are clear enough to review.

Check that your criteria relate to the job and consider how they could introduce bias. Don’t automatically reject candidates based only on a score, especially when information is missing or extraction is uncertain.

Claude can help you turn your screening requirements into a testable tool. You still verify credentials, clarify availability, assess job-relevant skills, and make the hiring decision. Use the structure to give your professional judgment more room to work.

Ready to build practical AI skills for your HR work? Explore Tixu.ai, a beginner-friendly AI learning platform.