Popular starting points

Professional CertificateBeginner

Generative AI

AI prompting, workflow automation, output evaluation, privacy, API integration and a practical AI project.

Talk to a counsellor about Generative AI

Fees, batch dates and timings are shared by a counsellor. No payment is needed to ask.

What you will cover
  • AI prompting
  • workflow automation
  • output evaluation
  • privacy
  • API integration
  • a practical AI project

Topic list from the institute's programme record.

Certification

Certificate terms for this programme are confirmed by the institute at admission. Any ADIVAS course-completion certificate is an institute certificate; it is not a university degree, an externally accredited qualification or a vendor certification (such as AWS or Microsoft) unless the institute states so in writing.

Curriculum 0.1 · Draft: written for ADIVAS and awaiting academic review and approval. Topics and weeks may change before your batch starts.

Your skill trail

You move through 4 stations. Each week builds one skill you can show, not just a topic you have heard.

  1. Understand the tool

    Know what generative AI is, how it fails and how to talk to it clearly.

    • Week 1 What generative AI is and is not
    • Week 2 Prompt quality

  2. Trust but verify

    Check output against sources and protect private data before any text is pasted.

    • Week 3 Verifying outputs
    • Week 4 Privacy, safety and fair use

  3. Workflows and APIs

    Build a draft-and-review workflow and call a model from a script safely.

    • Week 5 Workflow automation with a human in the loop
    • Week 6 Calling a model through an API

  4. Evaluate and deliver

    Test tools with evidence and complete a practical AI project with an honest report.

    • Week 7 Evaluating AI tools for real use
    • Week 8 Practical AI project and report

  5. Reviewed AI Reply Assistant

    Your project, reviewed by a trainer.

The ADIVAS mastery ladder: how each skill is recognised
  1. Explored You have seen the idea and tried a first example.
  2. Practising You can do it with hints and keep improving.
  3. Proficient You complete the weekly practical task correctly on your own.
  4. Proven You apply it in a project reviewed by your trainer.

Skill-level tracking is being added to the learner workspace. Your trainer already reviews weekly practical work.

What you will be able to do

  • Explain in plain words what a generative AI model does, what it cannot do, and why it can sound sure and still be wrong.
  • Write structured prompts with role, task, context, format and constraints, and improve them over several rounds.
  • Verify AI output against reliable sources and record what was checked, what was wrong and what was changed.
  • Protect personal, client and institute data when using AI tools, and choose what may or may not be pasted.
  • Design a repeatable workflow in which AI drafts and a person reviews, with a log of inputs and decisions.
  • Call a model through an API from a short script, handle errors, and keep keys out of the code.
  • Evaluate an AI tool with a small test set and decide, with evidence, whether it is useful for a real task.
  • Deliver a small practical AI project with a short report on benefits, limits and risks.

Who it suits

  • School Curious students in Classes 9-12 who can use a browser and want to learn to question AI answers rather than copy them.
  • College College students who want to use AI for study, reports and small automation while keeping their own work honest and checked.
  • Professional Office staff, shop owners, teachers and freelancers who want to save time on drafts and routine tasks without leaking data or publishing wrong facts.

Prerequisites: Comfortable typing and using a browser and email. Basic spreadsheet or document use. No programming needed for weeks 1-5; weeks 6-7 use short, provided Python scripts. Tools: A current general-purpose AI assistant (a free tier is enough), A second assistant from a different provider for comparison, Google Sheets or LibreOffice Calc, Python 3 and a code editor such as VS Code, A text file or sheet used as a prompt and verification log.

Weekly plan

Understand the tool Weeks 1–2

Week 1: What generative AI is and is not

Theory
  • Models predict likely text, they do not look facts up by default
  • Hallucination: fluent but invented answers
  • Training data, knowledge cut-off and bias
  • Where AI helps (drafts, summaries, ideas) and where it should not decide (grades, medical or legal advice)
Lab
Ask the same factual question about a local topic to two assistants. Record both answers in a table and mark each statement as correct, wrong or unverifiable.
Homework
Find one example of a confident wrong AI answer (your own test) and write three lines on why it happened.
Deliverable
A comparison table of two assistants with at least ten marked statements.
Weekly checkpoint
Short quiz on hallucination and limits, plus a trainer review of the comparison table.
You will be able to
Explain in plain words what a generative AI model does, what it cannot do, and why it can sound sure and still be wrong.

Week 2: Prompt quality

Theory
  • Role, task, context, format, constraints
  • Giving examples (few-shot) and asking for step-by-step reasoning visible in the answer
  • Iterating: change one thing at a time
  • Asking the model to ask you questions when the task is unclear
Lab
Take a weak prompt, 'write about Madurai', and improve it in four rounds into a prompt for a 150-word tourist note for school students. Save each round and the output.
Homework
Write three prompts for your own study or work task and test each twice, noting what changed.
Deliverable
A prompt log showing at least four versions of one prompt and a one-paragraph reflection.
Weekly checkpoint
Rank four prompts from weakest to strongest; rewrite one.
You will be able to
Write structured prompts with role, task, context, format and constraints, and improve them over several rounds.
Trust but verify Weeks 3–4

Week 3: Verifying outputs

Theory
  • Claims, numbers, names, dates and citations must be checked
  • Using primary and official sources
  • Fake references and made-up quotations
  • A verification log: claim, source, result, action
Lab
Ask an assistant for a 200-word summary of a short article you have printed. Check every claim against the article and mark errors; then ask it to cite sources for a general topic and try to find each one.
Homework
Verify five claims from an AI-written paragraph on any topic and fill the log.
Deliverable
A verification log with at least ten rows and the corrected paragraph.
Weekly checkpoint
Scenario: an AI answer includes a statistic and a book title; decide what to check first.
You will be able to
Verify AI output against reliable sources and record what was checked, what was wrong and what was changed.

Week 4: Privacy, safety and fair use

Theory
  • What counts as personal or confidential data (Aadhaar, phone, marks, client lists, passwords)
  • Anonymising: replace names with roles, remove numbers
  • Tool settings, data retention and the institute or employer rule
  • Copyright, citing AI help, academic honesty, deepfakes and consent
Lab
Given a mock customer complaint email with names, phone numbers and an order value, rewrite it into a safe version that still lets AI draft a polite reply; then draft and edit the reply.
Homework
Write a one-page personal rule sheet: what I will never paste into an AI tool, and what I will disclose.
Deliverable
An anonymised input, the AI draft, the edited final reply and the rule sheet.
Weekly checkpoint
Sort ten items into safe to paste, anonymise first, or never paste.
You will be able to
Protect personal, client and institute data when using AI tools, and choose what may or may not be pasted.
Workflows and APIs Weeks 5–6

Week 5: Workflow automation with a human in the loop

Theory
  • Choosing tasks: repetitive, low risk, easy to check
  • Draft, review, approve, send: where the person decides
  • Templates, spreadsheets and form responses as inputs
  • Logging inputs, outputs and reviewer changes
Lab
Build a workflow for sorting twenty mock student queries into categories with suggested replies using a spreadsheet and an assistant; a reviewer approves each reply before it is marked ready.
Homework
Draw the workflow for a task from your own life or shop and mark where a human must decide.
Deliverable
A workflow diagram plus the completed spreadsheet with reviewer column.
Weekly checkpoint
Identify the missing review step in a given workflow and justify it.
You will be able to
Design a repeatable workflow in which AI drafts and a person reviews, with a log of inputs and decisions.

Week 6: Calling a model through an API

Theory
  • What an API request and response are (JSON)
  • Keys, environment variables and why a key is never typed into code or shared
  • Handling errors, limits and costs
  • Temperature, token limits and fixed output formats
Lab
Run a provided short Python script that sends a prompt and prints the reply. Move the key to an environment variable, add a try/except for failures, and ask for a JSON reply with two fields.
Homework
Modify the script so that it summarises three different pasted paragraphs and prints a table.
Deliverable
A working script with the key removed from the file and a short README.
Weekly checkpoint
Code-reading questions: find where the key is exposed; predict what happens on a network error.
You will be able to
Call a model through an API from a short script, handle errors, and keep keys out of the code.
Evaluate and deliver Weeks 7–8

Week 7: Evaluating AI tools for real use

Theory
  • Pick a real task and define a pass mark before testing
  • A small test set: typical, hard and trick cases
  • Measuring accuracy, time saved, review effort and cost
  • Comparing tools fairly and recording failures
  • When the answer is: do not use AI for this
Lab
Create a test set of fifteen items for a task such as translating notices to Tamil or classifying feedback. Run two tools, score every answer, and compute accuracy and time saved.
Homework
Write a one-page recommendation: use, use with review, or do not use.
Deliverable
A scored test set with a results table and recommendation.
Weekly checkpoint
Interpret a results table and decide whether the tool passes the stated threshold.
You will be able to
Evaluate an AI tool with a small test set and decide, with evidence, whether it is useful for a real task.

Week 8: Practical AI project and report

Theory
  • Scoping a small project
  • Writing benefits, limits and risks honestly
  • Presenting the log, tests and human review evidence
Lab
Build and demonstrate your project, for example a study-notes helper, a shop product-description assistant or a feedback sorter, with prompts, verification, privacy handling and evaluation included.
Homework
Finish the report and rehearse a five-minute demo with one failure case shown.
Deliverable
A working project, a two-page report and a five-minute demo.
Weekly checkpoint
Project demo and viva using the rubric.
You will be able to
Write structured prompts with role, task, context, format and constraints, and improve them over several rounds. Verify AI output against reliable sources and record what was checked, what was wrong and what was changed. Protect personal, client and institute data when using AI tools, and choose what may or may not be pasted. Design a repeatable workflow in which AI drafts and a person reviews, with a log of inputs and decisions. Evaluate an AI tool with a small test set and decide, with evidence, whether it is useful for a real task. Deliver a small practical AI project with a short report on benefits, limits and risks.

Sample lessons

Week 2

Writing a prompt that gets a usable answer

A model answers what you ask, using the words you gave it. A vague prompt gets a vague, generic answer. A good prompt states who the answer is for, what task to do, what facts to use, how long and in what format, and what to avoid. You then read the result, change one thing and try again. This is iteration, not magic, and it saves time only when you also check the result.

  1. 1Role and audience
  2. 2Task and facts
  3. 3Format and limits
  4. 4Run and read
  5. 5Change one thing
Good prompts are improved in small steps.

Step by step

  1. State the role and audience, for example a teacher writing for Class 8 students.
  2. Describe the task in one clear sentence.
  3. Paste the facts the answer must use, so the model does not invent them.
  4. Set the format and length, such as five bullet points under 80 words.
  5. Add limits, such as use simple words and do not add facts that are not given.
  6. Run it, mark what is wrong, change one item and run again.

In real life

A shop owner in Madurai wants a WhatsApp message announcing a Pongal discount. A weak prompt gives a generic poster text; a prompt with the real discount, dates, shop name and a 40-word limit gives text that needs only a quick check.

Worked example

Problem. Improve this prompt: 'Write about water.' The task is a 100-word science note for Class 6 students on why we should save water.

Solution. Role and audience: You are a science teacher writing for Class 6. Task: write a note on why we should save water. Facts: Use only these points - only a small share of Earth's water is fresh and usable; wasted tap water adds up; fix leaks. Format: 100 words, simple sentences, one closing action line. Constraint: do not add numbers that I have not given. The answer is now specific, short and checkable.

Common mistakes

  • Asking for facts without giving a source and then trusting the reply.
  • Changing five things at once so you cannot tell what improved the answer.
  • Accepting the first answer without reading it.

Try it yourself

  • Rewrite 'make a poster text' for a real school event with date, place and audience.
  • Take one of your prompts and add a format and a limit; compare the two outputs and note differences.
  • Ask the assistant what extra information it needs for a task, answer it, and compare the result.

Week 3

Checking AI output before you use it

A language model produces text that sounds right because it follows common patterns, not because it has checked the facts. It may invent a number, a book, a law or a quotation. Your job is to find the claims that could hurt if wrong, such as names, dates, amounts and rules, and check each against a reliable source. Keep a log so that the check is visible and repeatable.

ClaimRisk if wrongSource checkedResult
UPI launched in 2005HighNational payments body websiteWrong, corrected
Fee is Rs 2,500HighInstitute fee sheetCorrect
A famous quoteMediumOriginal bookNot found, removed
A verification log makes your check visible.

Step by step

  1. Highlight every claim in the answer: numbers, names, dates, rules and quotations.
  2. Rank them by risk: what happens if this is wrong?
  3. Open an official or primary source for each high-risk claim.
  4. Mark each as correct, wrong or not found, and record the source address and date.
  5. Correct the text, and if the model was wrong, note the type of error.
  6. Only then share or submit the text.

In real life

A student asks an assistant for the last date of a scholarship and the amount. The reply gives a confident date. The student checks the official notice and finds a different date; submitting on the wrong date would have cost the application.

Worked example

Problem. An AI paragraph says: 'The Indian UPI system processes over 50 billion rupees in a minute and was launched in 2005 by a private company.' How do you check it?

Solution. Split into claims: (1) volume per minute, (2) launch year, (3) who launched. Each is checkable and high risk. Look up the official payments authority and central bank pages. You will find that UPI was launched in 2016 by the national payments body, so claims 2 and 3 are wrong; and the volume claim cannot be confirmed from the answer alone, so it is removed or restated with a sourced figure. Log each row with source and action.

Common mistakes

  • Asking the same AI whether its own answer is correct and treating that as a check.
  • Checking only the easy claims and skipping numbers and quotations.
  • Trusting a reference list that looks real without opening the sources.

Try it yourself

  • Take an AI paragraph on a topic you know and mark every wrong or unverifiable claim.
  • Ask for three sources on a local history topic and try to open each; report how many exist.
  • Rewrite an AI answer so that every number carries its source.

Try the questions

The same skills are asked at three levels: School (Classes 6–12), College and Professional. Switch the level and the language (English or Tamil), and say how sure you are before you check: knowing what you are unsure about is part of learning.

Classes 6–12: everyday examples, no prior knowledge. Practice only. Nothing here is graded or saved.

Question 1 / 4 · Recognise AI hallucination

A student asks an AI for the name of a book on Madurai history. The AI gives a title and author, but no library has it. What most likely happened?

A student asks an AI for the name of a book on Madurai history. The AI gives a title and author, but no library has it. What most likely happened?
How sure are you?

Projects

Reviewed AI Reply Assistant

A small coaching centre in Madurai gets many student and parent queries by message. Build a draft-and-review workflow that uses an AI assistant to prepare replies from a short fact sheet, while keeping private details out of the tool and recording every check.

Tasks

  1. Write a fact sheet of fees, timings and rules with 8 facts and use it as the only source in the prompt.
  2. Anonymise twelve mock queries before sending them to the assistant.
  3. Produce draft replies and verify each against the fact sheet, logging errors.
  4. Edit and approve the replies, and record time taken versus writing from scratch.
  5. Write a half-page note on risks and when a person must answer instead.

Sample input

Query: 'Hi, I am Priya, mobile 9xxxxxxxxx. What is the fee and batch time for the weekend course?'

Expected output

Anonymised query, verified reply that matches the fact sheet, a log of at least twelve rows, and a time comparison.

How it is assessed

CriterionWeight
Prompt quality and use of source facts20%
Privacy handling20%
Verification log and corrections25%
Workflow with human review20%
Honest report of limits15%

Stretch challenge: Add a script that reads queries from a CSV and drafts replies through an API, still requiring manual approval.

How you are assessed

Each week: five short questions (concept, scenario and one diagram or table question), a practical task recorded in the prompt and verification log, and a two-line reflection on what the AI got wrong.

Practical work. Learners are assessed on their own logs, workflows and scripts: prompt versions, verification tables, anonymised inputs, a working API script without exposed keys, and a scored tool test.

What trainers look forWeight
Prompt quality and iteration20%
Verification and correction of output25%
Privacy and responsible use20%
Working workflow or script20%
Evaluation evidence and honest reporting15%

Where this can lead

  • Prepares you to use AI tools carefully for drafting, research support and routine office tasks.
  • Prepares you for junior roles that involve content support, data entry automation or AI-assisted operations with human review.
  • Builds the base for further study in programming, data analysis or applied AI.

These describe skills you will build. ADIVAS does not guarantee jobs, placement or salary.