TA Pre-Assessment — Start

Describe one specific moment where AI got something wrong in your recruitment work — a bad shortlist, a wrong summary, a tone-deaf message, an irrelevant Boolean. What happened, and how did you catch it?

Which of these have you personally used in the last 30 days for recruitment work? (Select all that apply)

You’re hiring a Staff Data Engineer for a well-funded healthtech startup in Bengaluru (Series C, building India’s largest diagnostics data platform). Three recruiters have already failed to get this person to reply.

Target candidate — public profile info:

• Ananya Krishnan · Staff Data Engineer @ Razorpay, Bengaluru · 9 years
• Previously: Swiggy (data platform team during hypergrowth)
• LinkedIn activity: recently posted about migrating Razorpay’s batch pipelines to streaming (“3 things I’d do differently”); comments actively on data-mesh debates; shared a post about mentoring women in data engineering
• GitHub: maintains a moderately popular open-source dbt utility
• Not “open to work.” Gets recruiter InMails daily.

Your role’s honest selling points: first Staff-level data hire, green-field architecture ownership, healthcare data at population scale, strong funding, equity meaningful at this stage. Honest weaknesses: brand is unknown vs Razorpay, cash comp roughly equal (no big jump), team is small (would be 1 of 4 initially).

Task — submit all three parts: (a) The prompt: Write the complete prompt you’d give your AI tool to draft this outreach — including how you feed it her profile signals and your role’s honest positioning. (b) The message: The final outreach message you would actually send (under 120 words — she doesn’t read long InMails). (c) The edit log: 2–4 lines on what you changed from the AI’s draft and why — or, if you’d write this one fully by hand, defend that call.

Your delivery head messages you:

“Client escalation. The Providence account needs 40 shortlisted profiles for the analytics ramp by Monday 9 AM. I’ve got us access to that new AI screening tool on trial. Plan: upload all 600 applications from the job posting tonight, let the tool auto-reject the bottom 400, auto-send them rejection emails so the pipeline is clean, and we manually look at the top 200 over the weekend. Also — upload the interview transcripts from last quarter’s analytics hires so the tool can ‘learn our bar.’ It’s just a trial license on the vendor’s cloud, we’re not paying, so no procurement or infosec review needed. Confirm and start tonight.”

Task: Write your actual reply to your delivery head (the message you would really send), followed by a short bracketed note [for the assessor] listing every problem you spotted in this plan and your alternative plan for hitting Monday’s deadline.

Rubric (20):

• Problems spotted — 10 marks (2 each, max 10): (1) auto-rejecting 400 people via an unvalidated trial tool with zero human review; (2) auto-sending rejection emails at scale — irreversible candidate-experience and brand damage if the tool is wrong; (3) uploading 600 applicants’ personal data to an unvetted vendor cloud with no infosec/procurement review — DPDP consent and purpose-limitation violation; (4) uploading interview transcripts — highly sensitive assessment data of identifiable past candidates/employees into a third-party training loop; (5) “free trial = no review needed” fallacy — data exposure is identical regardless of license cost.
• The reply itself — 6 marks: 2 for tone (constructive, urgency-acknowledging — not preachy refusal or silent compliance); 2 for offering a deadline-viable alternative (use the tool to rank not reject, human-review the bottom band in batches, anonymise/strip PII if any upload proceeds, hold all rejection comms until validated, split the 600 across the team over the weekend); 2 for escalating the data question to the right owner (infosec/DPO) without stalling the delivery.
• Judgment quality — 4 marks: recognises the core principle: AI can compress the work, but irreversible actions (rejection, data exposure) need validation first; deadline pressure is exactly when this discipline matters.

Your ATS’s AI generated this candidate summary for the hiring manager. Below it are the verified facts from the candidate’s actual resume. Find every error in the AI summary. List each one and state what the resume actually says.

AI-GENERATED SUMMARY: “Rahul Deshpande is a seasoned engineering leader with 11 years of experience, currently serving as Engineering Manager at Flipkart, where he leads a 25-member platform team. Previously at Paytm for 4 years, he drove the migration of their payments stack to microservices, cutting transaction latency by 60%. He holds a B.Tech from IIT Bombay and is AWS Solutions Architect Professional certified. He has managed teams through two successful product launches and is an active open-source contributor. Notice period: 30 days.”

VERIFIED RESUME FACTS:

• Total experience: 9 years 2 months
• Current: Senior Engineering Manager, Myntra (Flipkart group company, but Myntra) — leads a 25-member platform team ✓
• Previous: Paytm, 2 years 8 months — worked on the microservices migration as one of three tech leads; resume claims latency improvement of 35%
• Education: B.Tech, COEP Pune (resume mentions he attended a certification program at IIT Bombay)
• Certification: AWS Solutions Architect Associate (not Professional)
• Two product launches ✓ · Open-source contributor ✓
• Notice period: 60 days, negotiable

Task: List every discrepancy you can find. For each: what the AI said → what is actually true. Then answer in one line: what would have happened if this summary had gone to the hiring manager unchecked?

Requisition brief from the hiring manager (verbatim, as received): “Need a DevOps person. Kubernetes is a must, AWS preferred. Someone who’s handled production incidents, not just built pipelines. 6–9 years. Bangalore, hybrid, 3 days. Budget max 45 LPA. Don’t send me people from service companies who’ve only done support tickets. Need to close in 4 weeks.”

Task: Write the exact, complete prompt you would give an AI tool to generate your LinkedIn sourcing strategy for this role — search strings, target companies, and title variants. Write the prompt itself, ready to paste into the tool. Do not describe what you would do — write it.

Paste or reconstruct, as closely as you can remember, one actual prompt you wrote for a recruitment task in the last month. Then tell us in one line: did the output work, and what did you change before using it?

A recruiter on your team prompted an AI tool: Prompt: “Write a rejection email for a candidate.” AI output: “Dear Candidate, Thank you for your interest in the position. After careful consideration, we regret to inform you that we have decided to move forward with other candidates whose qualifications more closely match our requirements. We wish you the best in your future endeavors. Best regards, The Hiring Team.”

Context the recruiter had but didn’t use: The candidate is Meera, a final-round candidate for a Product Manager role who spent 6+ hours across 4 interview rounds over 5 weeks. The panel genuinely liked her; she narrowly lost to an internal candidate. The client (your key account) wants her warm for a similar role opening next quarter. She has 12k LinkedIn followers in the product community.

Task (two parts): (a) In 2–3 lines, diagnose why this output is not just generic but commercially damaging in this specific situation. (b) Rewrite the complete prompt so the AI produces the message this moment actually needs. Write the full prompt, ready to use.

Your new AI screening tool scored 8 applicants for a Senior QA Automation Engineer role (Selenium, API testing, CI/CD, 6+ yrs). You have 15 minutes before your shortlist call. Here is the tool’s output:

Candidate	Yrs exp	Skills match (from resume)	Career note	AI Score
Arjun M	7	Selenium, REST Assured, Jenkins, Docker	Continuous employment	91
Kavitha R	8	Selenium, API testing, GitLab CI, led 4-member team	11-month break, 2022 (maternity)	58
Dinesh K	6	Selenium, Postman, basic CI	Continuous employment	84
Fatima S	9	Selenium, REST Assured, Jenkins, mentors juniors	8-month break, 2021 (parent care)	61
Rohit T	6.5	Cypress, API testing, CircleCI	Continuous employment	88
Priya N	7.5	Selenium, API testing, Jenkins, AWS	9-month break, 2023 (upskilling — cloud certs earned)	63
Sandeep V	6	Selenium only, no API testing on resume	Continuous employment	79
Meghna J	10	Full stack of required skills, conference speaker	7-month break, 2020	66

Task: (a) Which candidates do you take to the shortlist call, and in what order? One line of reasoning each. (b) What pattern do you see in this tool’s scoring? Be specific — cite the numbers. (c) What do you do about the tool — today, and this quarter?