About Handpick
How we read applications, and how we keep it fair
Handpick helps the Help Desk hiring team read every applicant carefully and respond to each of them warmly. It organizes, summarizes, and suggests. It never decides. People make every hiring decision, and every email a student receives is sent by a person on this team.
Where the information comes from
- Handshake: the student's application to the OIT Help Desk Student Consultant posting and the resume they attached. Handshake also shows graduation date, work authorization, and Work-Study eligibility; we display those, but they are never scored (see below).
- The HD application: our short Google Form, which students complete after applying.
- Interview availability: the self-scheduling page offers only open times inside the team's interview hours, skipping UCI holidays, anything on the recruitment Google Calendar, and the interviewer's Microsoft 365 appointments that show as Busy. Only start and end times are read; meeting details are not stored.
- Profile photos: shown only so the team can put a face to a name after meeting someone. Photos are never sent to the AI and never affect anything.
The preference score (for example, "81 pref")
The job is people-first. It does not require a technical or computer science background, so the rubric rewards the things that actually make a great Help Desk consultant. For each applicant, Claude (an AI model from Anthropic, running privately in UCI's Amazon Web Services account) reads the de-identified resume and rates six qualities from 1 to 5, using only evidence on the page:
The math. Each rating is turned into a fraction of 5, multiplied by its weight, and added up, giving a number from 20 to 100:
preference = 100 × (0.30 × customer/5 + 0.20 × communication/5 + 0.20 × availability/5 + 0.15 × reliability/5 + 0.10 × tech/5 + 0.05 × attitude/5)
Example. Ratings of 5, 4, 3, 4, 3, 5 give 100 × (0.30 + 0.16 + 0.12 + 0.12 + 0.06 + 0.05) = 81.
- Tiers: Strong is 75 and up, Solid is 55 to 74, and Worth a look is below 55. Every tier is a positive label; nobody is marked as rejected.
- Neutral when unknown: resumes rarely list availability, and a first-year student may have little work history. When something simply is not shown, it gets the neutral 3, not a low score. Missing evidence is not negative evidence.
- Tech is deliberately light: basic comfort with computers and apps earns full marks. Deep engineering experience earns no extra credit, because this role is about people.
The HD application read
The HD application is open book, so it is not graded like an exam. Claude gives a general, generous sense of how someone did: effort and completeness, clear writing, sensible troubleshooting instincts (asking questions, trying simple steps, knowing when to escalate), and how they would come across to a stressed customer. The result is one of Excellent, Strong, Solid, or Needs a closer look, with a short summary, what they did well, and gentle follow-up questions for the interview. Answers to identifying questions (name, email, phone, student ID, address) are removed before the read.
The HD application read is shown next to the preference score, not blended into it, so reviewers can see each on its own terms.
What never counts
These are never sent to the AI and never affect a score or grade:
- Name, photo, email, phone, and profile links
- Work authorization or visa sponsorship status
- Work-Study eligibility
- Graduation date or class year
- Gender, race, ethnicity, national origin, citizenship
- Religion, age, disability, sexual orientation, family status
- Which cultural, faith, or identity groups someone belongs to
- Prestige of an employer, school, or brand
Personal details are stripped from the resume before the AI sees it, and the AI is also instructed, every time, not to infer or weigh any of the above. Clubs and jobs count for what the person did (serving customers, leading, organizing, teaching), never for which group it was: a cultural club officer and a chess club officer who did the same work get the same credit. Clear writing is valued, but non-native English phrasing that is still clear is not penalized.
About the filters. Handpick can filter by Work-Study, work authorization, and graduation year because those are practical facts the team sometimes needs, for example Work-Study funding. Filters are for organizing, not screening. Work authorization or sponsorship status must not be used to rule out a student for this on-campus job; many students on visas can work on campus. Bring eligibility questions to HR.
Safeguards
- Nobody is auto-rejected. Every applicant stays visible, and anyone can be advanced regardless of score.
- A person sends every email. Each email is previewed and sent by a named team member. Handpick starts in test mode, where every send goes to the reviewer instead of a student.
- Everyone sees the same times. Students book their own interview from the same open slots, so nobody's chances depend on how quickly they answer an email or how they word their availability.
- Everyone gets the same words. Lifecycle emails come from shared templates, so every applicant receives the same information and the same warmth.
- No spam. Every email is logged per applicant. The bulk reminder skips anyone who submitted, was already reminded, or was asked in the last three days, and it can never send twice to the same person.
- A clear record. Who sent what, and when, is kept with each applicant and in the conversation history.
Using the score well
- Use the score to decide reading order, not outcomes. Read the resume and the HD application yourself before deciding.
- AI can be wrong. It can misread an unusual resume layout, miss experience described in unfamiliar terms, or overlook availability that was never written down. If a read seems off, trust your own reading.
- Thin resumes are common for newer students. The interview is where they get to show who they are.
- Use the suggested interview questions and follow-ups as a starting point, and ask every candidate the core questions so comparisons are fair.
Your data
Applicant data stays in UCI's AWS account, encrypted, and is visible only to Help Desk admins and supervisors who sign in with UCI credentials. The AI runs through Amazon Bedrock, which does not use this data to train models. Applicant information will be deleted after the hiring cycle ends.
Questions or concerns about fairness? Contact Mike Caban or write oithdrecruitment@uci.edu. Scoring model: Claude Sonnet 4.6 on Amazon Bedrock. Method last reviewed September 24, 2026.