GA3 Made Simple
Open exam
Choose a question

Your student field guide

Clear steps.
Less second-guessing.

Pick a question, follow the steps, and save in the exam portal. The submission links and the file limits are right where you need them.

Both buttons use an attempt. Q1 allows 3 attempts per 24 hours. Q2–Q4 allow 10 per question per 24 hours. Keep the final attempt for Save; for Q2–Q4, only your latest Save counts.

Q1Number guessingOptional exercise · no marks Q2Invoice agentCopy a link, Check, then Save Q3Mailroom agentUse the exact mailroom endpoint Q4Incident agentSubmit the incident-service link Q5Humans Only game1 bonus mark for attempting · see the email Q6Build your 3D faceMeshy 7 Pro first · free backup · GLB Q7Create a morph videoKling setup, prompt, and compression Q8Rank the face pairsAll ten pairs, easiest to hardest
Backup solver

Credits: GT Indian. Follow the backup site’s question-specific submission instructions.

Open backup solver

Question 1

Guess your number

Optional0 marks

A small number-guessing exercise. It does not contribute marks, so you do not need to solve it to continue.

  1. Open Q1 in the exam portal.

    Enter a whole number from 0 to 10,000.

  2. Use the feedback if you choose to play.

    Check tells you whether your guess is below, above, or equal to the target.

  3. Keep an attempt for Save.

    There are only 3 attempts per 24 hours. Check and Save both use one.

You can move on to the marked questions without completing this exercise.

Question 2 · Invoice / A2A

Submit the invoice agent

The grader sends invoice cases to the hosted solver. You submit its URL; you do not need your own API key or deployment.

https://tds-ga3-qult.onrender.com/
Backup solver

Credits: GT Indian. Follow the backup site’s question-specific submission instructions.

Open backup solver
  1. Hard refresh the exam portal first.

    In Chrome/Edge on Windows, press Ctrl + Shift + R while you are on the exam page. Chrome on Mac uses Cmd + Shift + R.

  2. Paste the URL directly into Q2’s answer box.

    Use the link above, with no extra words, token, or query string.

  3. Click Check and read the feedback.

    Check your remaining attempts before clicking. The observed Q2 scores reported for this service are 3.9–3.95/4; a particular score is not guaranteed.

  4. Save the answer you want counted.

    Save also uses an attempt and may run fresh verification. Read the saved feedback.

Avoid pressing Save repeatedly. Check and Save both consume attempts, and only the latest Save counts. Hard refresh clears stale portal presentation; it does not reset attempts or guarantee a stable score. Fresh audits can still differ.

Do not use all attempts on Check. The final attempt is needed for Save. A checked answer is not automatically a saved answer.

Question 3 · Mailroom

Use the mailroom link

This question needs the mailroom endpoint, not just the service homepage.

https://tds-ga3-qult.onrender.com/v1/mailroom/actions
Backup solver

Credits: GT Indian. Follow the backup site’s question-specific submission instructions.

Open backup solver
  1. Hard refresh the exam portal.

    Use Ctrl + Shift + R in Chrome/Edge on Windows, or Cmd + Shift + R in Chrome on Mac.

  2. Copy the complete endpoint into Q3.

    Keep /v1/mailroom/actions at the end. You do not need to paste an API key.

  3. Check, review, then Save.

    The grader handles the case and receipt requests. The service has returned full marks in reported tests, but each verification still needs to succeed.

Check and Save share the limit. Keep the last attempt for Save. Refreshing the portal does not reset the allowance; repeated Save clicks can replace your previous saved result.

Do not replace this endpoint with the homepage URL. Submit the complete mailroom URL shown above.

Question 4 · Incident response

Submit the incident agent

The grader sends incidents, diagnostic results, and approvals to the service. Submit the base URL below.

https://tds-ga3-qult.onrender.com/
Backup solver

Credits: GT Indian. Follow the backup site’s question-specific submission instructions.

Open backup solver
  1. Hard refresh the exam portal.

    Use Ctrl + Shift + R in Chrome/Edge on Windows, or Cmd + Shift + R in Chrome on Mac.

  2. Paste the base URL into Q4.

    You do not need to append /v2/incidents; the grader calls that route itself.

  3. Check and review the actual feedback.

    The service has returned full marks in reported tests. A new Check or Save can use a different audit incident, so results can differ.

  4. Save once you are ready.

    Keep an attempt available, and inspect the saved result rather than relying only on an earlier Check.

Repeated Save clicks consume attempts and can produce new audit results. A hard refresh helps with stale portal UI; it does not fix every server error, reset limits, or guarantee full marks.

If feedback reports a transport error, read that error first. Do not keep spending attempts without a reason to expect a different outcome.

Question 5 · Humans Only

Read the bonus note first

The drawing-and-guessing game carries a separate participation bonus, not regular question marks.

Q5 had a submission bug. The professor’s email to everyone announced 1 bonus mark for students who attempted the game.

The game, in simple terms

  1. Open your personal game desk from Q5.

    Use your own sign-in link. Keep that personal link private.

  2. Drawing stage: draw three assigned words.

    You get 60 seconds per drawing. Do not put letters or numbers into the drawing.

  3. Guessing stage: answer 15 drawings.

    Each is shown for 30 seconds with ten answer options. Follow any Google identity confirmation required by the game.

  4. Follow the professor’s submission workaround.

    Do not assume that repeatedly checking a broken result will repair it or change the bonus rule.

StagePublished window · IST
Drawings21 Oct, 12:00 PM to 24 Oct 2026, 8:00 AM
Guessing22 Oct, 12:00 PM to 25 Oct 2026, 12:00 PM
Published Check/Save deadline31 Oct 2026, 11:59 PM

As of this guide’s 29 October review, the drawing and guessing windows have ended. The steps above describe the game flow; they do not promise you can start a new attempt after the window closes. A newer course announcement takes precedence.

Question 6 · Image to 3D

Make your assigned face in 3D

Choose the higher-quality Pro route or the simplest free route. Only the selected method’s steps are shown; both use the same compression and deployment steps below.

Submit a valid GLB (glTF 2.0) containing a mesh, at most 2,000,000 bytes. The model must resemble your assigned reference; a small valid file alone does not prove likeness.

Meshy 7 Pro

MeshyGrab is the third-party downloader used in the author’s workflow. Use it only for assets you are authorized to export; it does not grant paid-plan export permission.

Get MeshyGrab extension

Only 2 extension download attempts were reported in the tested workflow. Finalize and preview your model before using one.

  1. Upload your reference to Meshy.

    Open Image to 3D and choose Meshy 7 Pro if your account offers it.

  2. Generate and inspect the model.

    Check the face from the front and sides. Regenerate before downloading if the likeness is poor.

  3. Add the texture.

    Click Texture / Generate Texture after the mesh appears. This may look like a large green texture tile.

  4. Download a GLB through a permitted export route.

    Use the official export where available, or MeshyGrab for an authorized export. Your final file must be GLB.

  5. Compress, preview, and upload.

    Follow the shared compression and GitHub CDN steps below.

Open Meshy

Meshy 6 Lite

The simplest free route. No extension is needed, but the likeness may be rougher than the tested Pro result.

  1. Upload your reference to Meshy.

    Open Image to 3D and select Meshy 6 Lite.

  2. Generate and check the face.

    Inspect the front and sides before spending time on the texture.

  3. Generate the texture.

    Click Texture / Generate Texture after the model appears.

  4. Export as GLB.

    Use Meshy’s official download option and check your remaining free allowance.

  5. Compress, preview, and upload.

    Use the same compression and GitHub CDN steps below.

Open Meshy
If your model looks unlike the reference

Try a cleaner image preparation with GPT, Gemini, or another image tool. Supply the assigned reference and request a 3D-friendly version that preserves the person’s face, hair, proportions, and identity.

Do not invent a different face. Keep the head clear, use consistent lighting and a simple background, then compare the prepared image against the original before generating again.

Optional image-preparation prompt
Use the attached assigned reference image as the identity source. Prepare a clean image suitable for image-to-3D generation. Preserve the same person’s facial proportions, features, hair and overall appearance. Keep the full head visible, use even lighting and a simple background, and do not beautify the face or invent missing identity details. The reference likeness matters more than a polished-looking replacement.
GLB compression request
Help me reduce this GLB to below 2,000,000 bytes while preserving the assigned face’s geometry and texture as much as possible. Keep the output as a valid glTF 2.0 GLB containing a mesh. Remove unused data and reduce texture size or mesh complexity carefully. Provide the compressed file if your tools support that; otherwise give reproducible commands. Report the final byte size and explain how I can preview the model to check that the likeness and texture still work.

Preview the model after compression. Check gives 0.5 marks for access/validity; the other 7.5 marks depend on resemblance in offline review.

Deploy with GitHub CDN

Upload the final file to a public GitHub repository and serve it through jsDelivr. This gives you a direct file URL with browser/CORS access.

  1. Create a public repository for your files.

    For example, name it ga3-assets. Use your own GitHub account.

  2. Upload the compressed file.

    Choose Add file → Upload files, upload face.glb, and commit. Use a simple filename without spaces.

  3. Copy the latest commit hash.

    Open the commit and copy its full hash. This pins the CDN link to your uploaded version.

  4. Build the CDN URL below.

    Replace USERNAME, REPOSITORY, COMMIT, and FILENAME.glb with your own values.

  5. Open the link in a private browser window.

    It should directly download or play the file, without a login or viewer page. Keep the repository and file available for offline review.

  6. Paste the CDN URL into Q6, Check, then Save.

    Submit the file link, not the GitHub repository page or a /blob/ URL.

GitHub CDN URL template
https://cdn.jsdelivr.net/gh/USERNAME/REPOSITORY@COMMIT/FILENAME.glb

If you replace the file, use the new commit hash in your link. A link using a moving branch name can keep serving an older cached version.

Question 7 · Image to video

Make a smooth face morph

Start with your first assigned face and finish with your second. Generate the video, compress it, and submit a direct file link.

A browser-playable MP4 or WebM, strictly below 100,000 bytes, with positive duration and nonzero dimensions. The exam does not specify a target duration; 6–7 seconds is a generation suggestion.

Start in Kling AI

  1. Sign in and check your credits.

    Your account may show a 66-credit allowance, but offers and model access vary. Review the displayed generation cost before confirming.

  2. Open Video Generation.

    Select Kling 3.0 Omni if your account offers it. Use the starting settings below.

  3. Assign the first and last images correctly.

    Use explicit start/end-frame controls if available. Upload the first assigned image as the start and the second as the end. Two generic reference uploads alone may not enforce exact endpoints.

  4. Paste the prompt and replace the filenames.

    Replace [FIRST_IMAGE_FILENAME] and [SECOND_IMAGE_FILENAME] with your uploaded names. Keep the same order.

  5. Generate and inspect the result.

    Watch the full video, especially the first and last frames. Reject cuts, identity swaps, flicker, or a final face that does not match the second reference.

SettingSuggested starting value
ModelKling 3.0 Omni, if available
Resolution720p for generation
Aspect ratio1:1 or Auto; keep both references framed consistently
Duration6–7 seconds
AudioOff / remove it before final compression
Open Kling AI
Face-morph generation prompt
Use the FIRST uploaded image, [FIRST_IMAGE_FILENAME], as the starting frame and the SECOND uploaded image, [SECOND_IMAGE_FILENAME], as the ending reference.

Create one continuous close-up portrait video. Begin with the first person’s face exactly as shown in the first image. Over the full duration, slowly and smoothly transform that face into the person in the second image. Gradually blend the skin, eyes, nose, mouth, jawline and hair, so the final face clearly matches the second reference.

Keep the head centred, the camera fixed, and the framing, pose, body, background and lighting consistent wherever possible. Use a natural, lifelike, photorealistic transformation. No cuts, flicker, abrupt identity changes, sudden jumps or generated text overlays.

720p generation is not the final submission size. A 6–7 second download normally needs substantial compression to fit below 100,000 bytes. Remove audio and lower resolution/frame rate as needed, while keeping the face transition readable.

Video compression request
Help me compress this face-morph video into a browser-playable MP4 or WebM strictly below 100,000 bytes. Aim for about 90,000 bytes to leave container overhead margin. Remove audio and reduce resolution, frame rate and bitrate carefully. Preserve the smooth transition and recognisable first/last faces as much as possible. Provide the file if your tools support that; otherwise give reproducible FFmpeg commands. Report the final byte size and explain how to verify playback, duration and dimensions.

Before uploading, play the compressed copy from start to finish and verify it is below 100,000 bytes. Check gives 0.5 marks for a valid playable file; another 4 marks depend on classmates’ offline assessment.

Deploy with GitHub CDN

Upload the final file to a public GitHub repository and serve it through jsDelivr. This gives you a direct file URL with browser/CORS access.

  1. Create a public repository for your files.

    For example, name it ga3-assets. Use your own GitHub account.

  2. Upload the compressed file.

    Choose Add file → Upload files, upload morph.mp4, and commit. Use a simple filename without spaces.

  3. Copy the latest commit hash.

    Open the commit and copy its full hash. This pins the CDN link to your uploaded version.

  4. Build the CDN URL below.

    Replace USERNAME, REPOSITORY, COMMIT, and FILENAME.mp4 with your own values.

  5. Open the link in a private browser window.

    It should directly download or play the file, without a login or viewer page. Keep the repository and file available for offline review.

  6. Paste the CDN URL into Q7, Check, then Save.

    Submit the file link, not the GitHub repository page or a /blob/ URL.

GitHub CDN URL template
https://cdn.jsdelivr.net/gh/USERNAME/REPOSITORY@COMMIT/FILENAME.mp4

If you replace the file, use the new commit hash in your link. A link using a moving branch name can keep serving an older cached version.

Question 8 · Face-pair ranking

Rank all ten pairs

Place the pairs you think are easier to morph above the harder ones. Use your own assigned pairs; there is no universal answer order.

  1. Inspect every assigned pair.

    Compare all ten before deciding the order. Watch any supplied morph previews where available.

  2. Look for alignment, not attractiveness.

    Similar head pose, framing, expression, facial-feature positions, silhouette and lighting generally make a smooth morph easier. Large changes need more transformation.

  3. Order the pairs from easiest to hardest.

    Use the drag handles or up/down controls in the portal. Put the easiest at the top and the hardest at the bottom.

  4. Include pair numbers 1–10 exactly once.

    If you keep the initial order, use the portal’s confirmation button rather than leaving the question untouched.

  5. Check, review, and Save.

    You submit the ranking directly in the portal. No hosted URL is required.

Check awards 1 mark for a complete ranking. Another 4 marks depend on agreement with peers’ assessment of the actual morphs.

A friend’s ranking may refer to different assigned pairs. A complete ranking passes the completeness check; it does not guarantee the peer-scored marks.