Manipur International University Students: Getting Mentor Feedback Remotely on Your AI Work

To get useful mentor feedback remotely on AI work, send a short written request that states your goal, what you tried, one to three specific questions and a deadline, then attach a five-minute screen-recorded walkthrough. Log every comment in a tracker so each piece of advice turns into a dated action. This guide gives you all three tools, built for a 14-day internship.

Why asynchronous feedback on AI work fails so often

A mentor reviewing your work remotely cannot see your screen, your data folder or your thinking. They get a message, maybe a notebook link, and ten spare minutes. AI work makes this harder than most subjects, because a result depends on invisible things: which split of the data you used, which random seed, what the baseline was, and which version of the notebook produced the number you quote.

Most weak feedback traces back to one of four causes:

• The question is too broad. “Please review my project” invites a polite, generic reply.
• The work cannot be reproduced. If the mentor cannot run it or follow it, they can only comment on how it looks.
• The context is missing. The mentor does not know your deadline, your skill level or what you have already tried.
• Nothing happens afterwards. Comments get read, forgotten and never checked against the next version.

The tools below address each cause directly.

A 14-day plan with three feedback checkpoints

In a two-week internship you have time for about three review cycles, not ten. Plan them in advance so each one has a clear purpose. The free two-week online internship from CareerSwami is built around a 14-day window, so this schedule fits it; adjust the days if your own timetable differs.

Leave days 13 and 14 for applying the last round of comments. Never submit a request on the final day and expect to act on the answer.

Feedback-request template for AI projects

Copy this into an email or message. Keep the whole request under 200 words, and replace every bracket.

Subject: Feedback request - [project name] - checkpoint [1/2/3] - reply by [date]

Hello [Mentor name],

Goal: [one sentence: what the project predicts, classifies or generates, and for whom]
Stage: [problem plan / baseline / final]
What I did: [2-3 lines: data used, model tried, metric]
Current result: [one number with the metric name, and the baseline it is compared to]
Where I am unsure:
1. [specific question, e.g. "Is my train/test split valid given repeated users?"]
2. [specific question]
3. [optional]

Files: [link to notebook/repo, link to 5-minute video]
Time needed: about [10] minutes
Please reply by: [date, at least 24-48 hours ahead]

Thank you,
[Name, programme, track]

Rules that make the template work

• Limit yourself to three questions. Mentors answer short lists; they skim long ones.
• Quote a number with its comparison. “Accuracy 0.91” means little; “accuracy 0.91 against a majority-class baseline of 0.89” invites a real discussion.
• Share a view-only link and check it opens in a private browser window before sending.
• State the deadline honestly and give a gap of at least a day.
• Do not paste private or personal data into a message or a video. Use a public or anonymised dataset.

A five-minute video walkthrough format for asynchronous review

A short recording replaces the screen-share you cannot have. Use any free screen recorder, keep your voice clear, and stick to this running order.

Recording habits worth adopting

1. Write a three-line script first, so you do not ramble.
2. Restart the notebook and run all cells before recording, to confirm it works from top to bottom.
3. Zoom the browser so code is readable on a phone screen.
4. Add timestamps for each question in the message, so the mentor can jump straight to them.
5. Upload with link access limited to people who have the link, and confirm captions or audio are clear.

Feedback action tracker you can copy

A tracker converts comments into work. Keep it in a spreadsheet or a plain table in your README. Log each comment within a day of receiving it.

Use these category labels: Method, Evaluation, Data, Code quality, Communication, Scope. Use these statuses: Open, In progress, Done, Deferred, Declined. Declining is allowed if you write a one-line reason, which also shows mentors you are thinking rather than just obeying.

Worked example: one feedback cycle on a sentiment classifier

This example is illustrative. The numbers are invented to show the process, and no code was run for this article.

Asha is building a movie-review sentiment classifier. At checkpoint 2 she sends the request template with a walkthrough video. Her message says: “Current result: accuracy 0.93 against a majority-class baseline of 0.50. Question 1: is my split valid? Question 2: should I report F1 as well?”

The mentor replies in two lines: the vectoriser was fitted before the split, which can leak vocabulary information, and F1 is a good addition. Asha logs two tracker rows (F1 under Evaluation, F2 under Evaluation), refits the vectoriser on training data only, and re-runs. Her illustrative new accuracy is 0.90. She records both numbers in the tracker and adds one sentence to the README explaining the drop. At checkpoint 3 she links to those tracker rows, so the mentor can verify the changes in seconds.

The lesson is that the lower number is a better result, because it is honest, and the tracker proves she acted on advice.

Questions to put in a request, by type of AI work

Pick the line that fits your project rather than asking something vague.

• Classification: Is my metric right for this class balance? Could any feature leak the label?
• Regression or forecasting: Does my validation respect time order? What is a sensible naive baseline?
• NLP: Are my train and test texts truly separate? Which errors should I inspect by hand?
• Computer vision: Is augmentation applied only to training data? Are near-duplicate images spread across splits?
• Generative or LLM prompts: How should I judge output quality without a ground-truth label? Is my prompt set varied enough?
• Any project: Is my README enough for a stranger to reproduce this?

Manipur International University students: fitting remote feedback around your own academic structure

MIU states that it was established under the Manipur International University Act, 2018, and runs its own schools, including Engineering and Information Technology, rather than a network of affiliated colleges. If you study in one of those schools, your faculty may be a natural first reader of your AI work, but you should ask rather than assume.

Timing matters. The 2026-27 academic calendar shows the odd semester running from July to November 2026, with end-terms in November. If you are reading this in mid-November, you may be near exams, so plan your 14-day internship around them and avoid sending review requests during your own exam days. The calendar itself says it is subject to revision, so check the current version.

MIU also lists a Project Development Cell that supports student and faculty innovation projects with the National Innovation Foundation and DST. That is a research and innovation cell, not an internship rule, so do not treat it as a route for internship approval without asking. No internship or project-work regulation, and no working placement cell page, was verified, so questions about credit, evaluation or documentation need a written answer from the university.

Verification steps for MIU students using remote mentor feedback

1. Email your department or school office and ask whether an external online internship and its mentor feedback can be shown in any academic record, and in what form.
2. Ask in writing who your internal faculty contact is and whether they are willing to read your tracker and final README.
3. Check the current academic calendar and note your mid-term and end-term dates before setting checkpoint days.
4. Ask the Project Development Cell whether it offers any review or guidance for student AI projects, and what it requires from you.
5. Look in the Public Self Disclosure section for any regulation on internships or project work, and request the document by name if you find none.

Etiquette for mentor communication that keeps replies coming

• Send one consolidated message per checkpoint, not a stream of small ones.
• Thank the mentor in one line and report what changed, since people respond to visible progress.
• Respect time zones and weekends; ask for a reply window instead of an instant answer.
• If you disagree with a comment, say so politely and give the evidence.
• Never present mentor suggestions as entirely your own work; credit them in your README.

Pre-send checklist for every feedback request

1. The notebook runs top to bottom after a restart.
2. The subject line names the project, checkpoint and reply date.
3. No more than three questions, each tied to a file, cell or timestamp.
4. At least one number is shown with its baseline.
5. Both links open in a private browser window.
6. The tracker is updated with the previous round’s actions.
7. No personal data appears in files, charts or the video.

Frequently Asked Questions

How long should an AI feedback request to a remote mentor be?

Aim for under 200 words plus a link to your notebook and a five-minute video. The mentor should be able to understand your goal, your result and your questions in under two minutes of reading.

Should I send the video, the notebook, or both?

Send both. The notebook lets the mentor inspect details and the video explains your reasoning and shows where your doubts are. If you must choose, send the video with a read-only notebook link, because it gives context faster.

What if my mentor does not reply by my deadline?

Send a short, polite reminder one day after the date, restating the single most important question. Meanwhile, continue with the parts that do not depend on the answer, and note the pending item as Open in your tracker.

How many feedback rounds fit in a 14-day internship?

Three is realistic: an early plan review, a baseline review around the middle, and a final review near day 12. More rounds usually leave too little time to act on comments.

Is it fine to decline a mentor’s suggestion?

Yes, if you record a clear reason in the tracker, such as limited time or a conflict with your goal. Mark it Declined or Deferred and mention it in your limitations section.

What should I do if the feedback is vague, like “improve the model”?

Reply with a narrower question, for example: “Do you mean trying a different algorithm, tuning hyperparameters, or improving the features?” Offering options makes it easy for the mentor to answer in one line.

Can I show mentor feedback to my university faculty?

You can share your tracker and README with faculty if the mentor is comfortable, but ask the mentor first. Whether any of it counts towards your university record is a question for your department, and it should be asked in writing.

Does the free CareerSwami internship include a mentor feedback structure I can follow?

The confirmed offer is a free 14-day online internship with an offer letter and a completion certificate after successful completion. The checkpoints in this guide are a self-managed routine that you can apply inside that window, not a promised feature.

Related Artificial Intelligence guides

• JNU Students from Humanities and Social Sciences: Entering an AI Internship Without a CS Degree
• Adamas University Students: Building a Recommendation System Mini Project During an Internship
• CMR University Students: Presenting an AI Internship Project to Non-Technical Managers
• All Artificial Intelligence internship guides