← WhatTNext AI सर्टिफ़िकेशन परीक्षा गाइड

सर्टिफ़ाइड AI एसोसिएट

कोर्स और परीक्षा अभी अंग्रेज़ी में हैं। उनके आसपास के पेज हिंदी में हैं।

यह किसे प्रमाणित करता है

A Certified AI Associate uses generative AI to do real work better and safely. They design and iterate prompts for structured tasks; verify outputs against sources and rubrics; apply AI to documents, data, meetings and simple automations; work with an organisation's own documents through retrieval tools; follow data-protection rules for client and employee data; and measure whether AI actually helped. Typical holders are analysts, executives, teachers, marketers, HR, finance and operations staff, and freelancers who use AI every week.

किसे देनी चाहिए

सुझाया गया अनुभव

Foundational-level knowledge (the Certified AI Foundational course or equivalent) and about three months of regular AI use at work or study. No programming needed; you should be comfortable with spreadsheets and email.

परीक्षा

  • 90 मिनट में 40 अंक वाले सवाल
  • 40% ज्ञान, 60% व्यावहारिक क्षमता
  • एक व्यावहारिक कार्य, स्कोर का 10%, नतीजा परीक्षा ख़त्म होते ही
  • 4 अतिरिक्त परीक्षण सवाल जो गिने नहीं जाते, ताकि नए सवाल परखे जा सकें
  • एक जवाब, कई जवाब (सब सही हों तभी अंक) और क्रम वाले सवाल, कई असली प्रॉम्प्ट और AI जवाबों पर बने
  • नतीजा: एक तय पास मानक पर पास या फ़ेल, हर सेक्शन के लिए "मानक पूरा" या "सुधार चाहिए"
  • ऑनलाइन आँकी गई, एक बार में एक सवाल, शुरुआत में वैकल्पिक वेबकैम पहचान जाँच
कोर्स लीजिए $10 / ₹399

सेक्शन और भार

सेक्शनभारजाँचे जाने वाले कौशल
Prompt design and iteration20%
  • Apply prompt patterns — role, step-by-step, examples (few-shot), structured output, critique-and-revise — to a work task.
  • Specify an output format (table columns, headings, fields) that people or tools downstream can use directly.
  • Separate instructions from pasted source material with clear delimiters.
  • Diagnose why a prompt produced a weak output and choose the edit that fixes it.
  • Break a large task into a sequence of smaller prompts with checks between them.
  • Build a reusable prompt template with placeholders for a recurring task.
Evaluating and verifying AI output20%
  • Check an output against a rubric of accuracy, completeness, relevance, tone and format.
  • Trace each factual claim in a draft to a source and flag claims with no support.
  • Detect errors in AI-produced numbers, formulas and tables by recomputing samples.
  • Identify omissions and distortions when AI summarises a longer document or meeting.
  • Recognise biased or non-inclusive language in AI drafts and revise it.
  • Decide when an output needs review by a subject expert before use.
AI in everyday work: documents, data and communication15%
  • Use AI to draft, restructure and tailor documents, emails and slides for a named audience.
  • Turn meeting transcripts into decisions, owners and dates, and check them against the transcript.
  • Use AI to write and explain spreadsheet formulas, then test them on cases with known answers.
  • Ask AI to write and explain a simple script or query, and run it first on a copy or test data.
  • Clean and categorise text data, such as survey comments, with AI and spot-check a sample.
Automation, assistants and your own documents15%
  • Map a repeated task into trigger, steps, AI step, human check and output for a no-code automation.
  • Configure a custom assistant with instructions, knowledge files and clear limits.
  • Explain retrieval-augmented generation (RAG) and why it reduces, but does not remove, made-up answers.
  • Diagnose RAG failures: missing or outdated documents, the wrong passage retrieved, or an answer not supported by the passage.
  • Decide where a human approval step must sit in an automated workflow.
Data protection and governance at work20%
  • Classify information as public, internal, confidential or personal before using it with an AI tool.
  • Apply an organisation's AI acceptable-use policy to a given situation.
  • Explain Data Fiduciary duties under India's DPDP Act 2023: notice, consent, purpose limitation, security safeguards, breach notification and erasure.
  • Distinguish consumer AI tools from enterprise tools with contractual data protections, and choose between them.
  • Redact or pseudonymise personal data before AI processing.
  • Explain what the EU AI Act's AI-literacy article and India's AI Governance Guidelines expect from organisations that use AI.
  • Recognise confidentiality and intellectual-property risks in AI inputs and outputs.
Measuring value and deciding when not to use AI10%
  • Define a baseline and measure time saved, quality and rework for an AI-assisted task.
  • Include checking and correction time when estimating productivity gains.
  • Identify tasks where AI must not decide alone, such as high-stakes decisions about people, legal sign-off and safety-critical judgement.
  • Compare AI with a simpler non-AI solution, such as a template, formula or checklist, for a given task.

यह स्तर किन ढाँचों के अनुरूप है

हमने पाठ्यक्रम बनाने में इन सार्वजनिक ढाँचों का इस्तेमाल किया। ये इस सर्टिफ़िकेशन का समर्थन नहीं करते।

नमूना सवाल

ये शैली और स्तर के उदाहरण हैं। ये परीक्षा में नहीं आते।

1.

Your manager needs the top five complaint themes from 312 app-store comments, with how many comments mention each and one example quote per theme. You used the prompt below and got the output shown. Which change to the prompt most directly fixes the problem?

प्रॉम्प्टSummarise this customer feedback. [312 comments pasted here]
AI का जवाबCustomers have mixed feelings about the app. Many like the design, while some have concerns about delivery times and pricing. Overall, sentiment is positive with some room for improvement.
  1. Start the same prompt with "You are a world-class senior customer-insights analyst with twenty years of experience in consumer apps."
  2. Ask for a table of the top five complaint themes with columns Theme | Comments mentioning it | Verbatim quote, using only the pasted comments.
  3. Add "Please be very accurate and thorough this time, because this summary is going to senior management for an important budget decision next week."
  4. Ask for a longer, more detailed summary of at least 500 words that covers every single comment.

जवाब: B

The original prompt never stated the deliverable. Specifying the structure (themes, counts, quotes) and grounding it in the pasted comments fixes that. A role line or 'be accurate' doesn't tell the model what to produce, and length isn't the goal.

2.

Compare the AI's action list with the meeting transcript. Which two problems does the action list contain?

प्रॉम्प्टTranscript extract: Neha: So we agree the vendor demo moves to Thursday the 16th. Ravi: Yes. I'll send the invite. Kiran, can you get the security questionnaire back from them by Monday? Kiran: I can do Tuesday, not Monday. Neha: Tuesday is fine. Task: List the decisions and the action items, with owners and dates.
AI का जवाबDecisions: - Vendor demo moved to Thursday 16th. Actions: 1. Ravi — send the demo invite. 2. Kiran — get the security questionnaire back from the vendor by Monday. 3. Neha — approve the vendor contract after the demo.
  1. Kiran's due date is wrong: Tuesday was agreed, not Monday.
  2. Action 3 was never agreed in the transcript.
  3. The demo date is wrong; the transcript moves the demo to a different day of the week.
  4. Ravi did not offer to send the invite; Neha said she would send it after the meeting.
  5. The actions should be listed in alphabetical order of the owners' names.

जवाब: A, B

The summary repeated the first-proposed date instead of the agreed one and invented an action that nobody agreed. These are the two most common meeting-summary failures: distortion and invention.

3.

In your sheet, column B holds each order's region and column D its value (rows 2–500). You asked an assistant for the average order value in the North region and it gave the formula below. What is wrong with it?

AI का जवाब=AVERAGEIF(D2:D500,"North",B2:B500)
  1. The range to test and the range to average are swapped; it should be =AVERAGEIF(B2:B500,"North",D2:D500).
  2. Nothing is wrong; AVERAGEIF accepts its two ranges in either order and returns the same result.
  3. "North" must be written without quotation marks, or the spreadsheet will treat it as the name of a cell rather than as text.
  4. AVERAGEIF cannot use text criteria, so you must use SUMIF and divide by the total count of all orders instead.

जवाब: A

AVERAGEIF takes (range to test, criterion, range to average). The formula tests column D for 'North' and averages column B. Testing on a few rows with a known answer catches this kind of error.

4.

A company's HR assistant answers questions from the company's own documents using retrieval-augmented generation (RAG). It tells an employee they get 12 days of earned leave. The leave policy updated in April says 15, and the April policy is in the system. What is the most likely cause?

  1. The model's training data is older than April, so it always ignores any documents that are retrieved for it.
  2. The old policy is still in the knowledge base, and retrieval returned it instead of the April version.
  3. RAG systems can only read PDF files, and the April policy was saved as a Word document instead.
  4. The employee asked the question in Hindi, and RAG systems can only answer questions asked in English.

जवाब: B

RAG answers from whatever passages it retrieves. If superseded documents stay in the index, the assistant can quote them. Remove or mark old versions and show the source and its date with every answer.

5.

You are automating supplier-invoice handling with a no-code tool and an AI step. Put the steps in the right order.

  1. Approved entries are posted to the accounting system and the invoice is archived
  2. Trigger: a new invoice email arrives in the shared accounts inbox
  3. An accounts executive reviews the flagged items and approves or corrects them
  4. The AI step extracts supplier, invoice number, date, amount and GSTIN into fields
  5. An automatic rule matches amount and supplier against the purchase order and flags mismatches

जवाब: B, D, E, C, A

Trigger → extract → rule-based check → human review of exceptions → post. The human check comes before anything is written to the books.

6.

Under India's DPDP Act 2023, when a personal data breach occurs, whom must the Data Fiduciary inform?

  1. Only the Data Protection Board of India
  2. The Data Protection Board and each affected Data Principal
  3. Only affected Data Principals, and only if more than 1,000 people are affected
  4. Only the police, through a cyber-crime complaint filed within 30 days

जवाब: B

Section 8(6) requires intimation to the Board and to each affected Data Principal, in the form and manner the Rules prescribe. Separate sector or CERT-In reporting duties may also apply.

7.

A colleague at an accounting firm wants to paste a client's full payroll spreadsheet (names, PAN, bank account numbers, salaries) into a free consumer chatbot to find unusual payments. The firm's AI policy allows client data only in its approved enterprise tool. What should happen?

  1. Paste it into the free chatbot, then delete the file from the laptop's downloads folder so that no copy is left behind afterwards.
  2. Use the approved enterprise tool, and remove or mask identifiers the analysis doesn't need first.
  3. Ask the client for verbal permission on a phone call and then paste the full file into whichever chatbot is quickest.
  4. Paste only half of the rows at a time, so the chatbot never sees the complete payroll file in one chat session.

जवाब: B

Policy decides the tool, and data minimisation decides what goes in. Splitting the file or deleting a local copy doesn't change what the consumer service received.

8.

Before AI, each monthly client report took 60 minutes to draft and 10 minutes to review. With AI, drafting takes 20 minutes, but review and correction now take 25 minutes. How much time does AI really save per report?

  1. 40 minutes
  2. 25 minutes
  3. 15 minutes
  4. 45 minutes

जवाब: B

Before: 60 + 10 = 70 minutes. After: 20 + 25 = 45 minutes. Saving = 25 minutes (about 36%), not the 40 minutes the drafting step alone suggests. Count checking time.