Sir Padampat Singhania University Students: Turning an AI Internship into a Research Poster

Yes, a 14-day AI internship can become a credible research poster, provided you scope it as one small, well-posed machine learning question rather than a full project. A good poster states the question, the data, the method, one or two honest results and the limits, all readable in about three minutes. The template and checklist below show how to do it.

What a Two-Week ML Study Can and Cannot Support

Fourteen days is enough for a narrow comparison: one dataset, one task, two or three models, a clear metric. It is not enough for new-architecture claims, large-scale benchmarks or deployment statements. Conference poster sessions at student symposiums reward clarity and honesty about scope, so frame your work as a small study or preliminary experiment.

Good poster-sized questions look like this:

• Does a simple baseline (logistic regression) match a small neural network on this tabular dataset?
• How much does class rebalancing change recall on a minority class?
• Which of three text-vectorisation choices gives the best F1 on a short-text classification task?
• How does image augmentation affect accuracy when only a few hundred training images are available?

Weak questions are open-ended (“Using AI for healthcare”) or claim-heavy (“A better model than existing work”). If you cannot state your question in one sentence, narrow it before building anything.

A 14-Day Plan Built Backwards from the Poster

If you join the free two-week online internship at CareerSwami, which gives a free offer letter and a free completion certificate after successful completion, plan the poster from day one. Choose from its 26 tracks the one closest to your question, and keep a running log so the poster is mostly assembly at the end.

Poster Layout Template for a Small ML Study

A common format is A0 portrait or landscape, or A1 for smaller boards. Confirm the required size with the event organiser before designing. The layout below uses three columns in landscape and reads left to right, top to bottom.

Design rules that keep a poster readable

• Title around 80-100 pt, section headings 40-48 pt, body text no smaller than 24 pt on A0.
• Keep total text under roughly 300-400 words; let figures carry the argument.
• Use two colours plus neutral grey, and check that charts stay legible in greyscale.
• Label every axis, include units, and put the metric name in the chart title.
• Export figures as vector files (SVG or PDF) or at 300 dpi so they do not blur when printed.

Content Checklist for a Conference-Style ML Summary

Use this list before printing. Every item should be answerable by looking at the poster alone.

1. Question: Is the research question a single sentence with a measurable outcome?
2. Dataset: Are the name, source, licence, number of samples, class balance and split sizes shown?
3. Leakage control: Did you split before preprocessing steps that learn from data (scaling, vectorising, oversampling)?
4. Baseline: Is there a simple baseline, such as majority class or logistic regression, shown beside your model?
5. Metric choice: Is the metric justified? Accuracy alone misleads on imbalanced data; add precision, recall or F1.
6. Variability: Do you report results over several random seeds or folds, with a spread such as standard deviation?
7. Fair comparison: Did every model get the same data split and a similar tuning effort?
8. Error analysis: Do you show where and why the model fails?
9. Claims match evidence: Does each conclusion bullet point to a specific number or figure?
10. Limitations: Do you name small data, one dataset, limited tuning and no external validation where they apply?
11. Reproducibility: Are library versions, seeds and a link or QR code to notes or code included?
12. Credit: Are datasets, papers, tutorials and mentors acknowledged and cited?

Illustrative Example: A Small Text-Classification Poster Plan

The following is a hypothetical plan, not a record of experiments that were run, and the numbers are placeholders showing how to present results.

Title: Do Simple Features Match Embeddings for Short-Text Complaint Routing? A Small Comparison.

Question: On a public set of short customer-complaint texts, does TF-IDF with logistic regression perform within 2 F1 points of a small pretrained-embedding classifier?

Method: Fixed 70/15/15 split, five seeds, three models: majority-class baseline, TF-IDF with logistic regression, embeddings with logistic regression.

The results box would then follow this shape, with placeholder values to be replaced by your own measured numbers:

Conclusion style: “In this dataset, the embedding model improved macro-F1 by [X] points over TF-IDF, a gap [larger/smaller] than the seed-to-seed spread.” That wording reports a measured difference without claiming it generalises.

The limitations box could read: single dataset; English text only; limited hyperparameter tuning; no test of other embedding models.

Writing the Conference-Style Summary and Two-Minute Pitch

Many student symposiums ask for a 150-250 word abstract alongside the poster. Use a fixed five-sentence structure:

1. Context: one sentence on why the task matters.
2. Question: what you tested.
3. Method: data and models in one sentence.
4. Result: the main number with its spread.
5. Implication and limit: what it suggests and what it cannot show.

For the spoken pitch, rehearse this order: the question (15 seconds), the data and method (30 seconds), the key result (30 seconds), the limitation and what you would do next (30 seconds). Then stop and invite questions.

Questions to prepare for at the poster

• Why this metric, and why not accuracy?
• Could information have leaked from test data into training?
• How large is the difference compared with random-seed noise?
• What would change if the dataset were larger or from another source?
• What did the model get wrong, and why?

Honest Reporting: Mistakes That Weaken ML Posters

• Reporting one lucky run. Repeat with several seeds and show the spread.
• Tuning on the test set. Use validation data for choices; touch the test set once.
• Overstated titles. “Novel state-of-the-art approach” is not supported by a two-week comparison.
• Unreadable figures. Screenshots of notebook output with tiny text rarely survive printing.
• Missing data provenance. A poster that cannot say where the data came from loses trust at once.
• Copying tutorial code without credit. Cite the tutorial or library documentation you built on.

SPSU Students: Fitting an AI Poster into Your University’s Internship and Placement Framework

Sir Padampat Singhania University describes itself as established under a 2008 Rajasthan State Act, recognised by UGC through a notification dated 13 April 2009, with its campus at Bhatewar, Udaipur. Its FAQ states that an internship is part of the programme curriculum, but that page gives no duration, credits, semester or evaluation details. So do not assume a CareerSwami internship or a poster substitutes for any required university internship; treat it as additional, self-directed work unless your department says otherwise in writing.

Two other points matter for a poster. First, the university’s Training and Placement Cell rules for session 2022-23 say a graduating student can lose placement assistance for not completing the summer internship programme, project training or other academic requirements. These rules are from 2022, so check the current version. Completing your official requirements comes first; a poster is a bonus on your resume.

Second, the B.Tech CSE (Core) page lists a Data Science specialisation, and the university also lists a B.Tech CSE in AI and ML programme run with Xebia (based on a page excerpt, so confirm details). If you are in one of these, a small ML study can connect naturally with your coursework, but the subject-level syllabus was not available to verify, so ask your faculty which topics overlap before choosing a question.

Verification steps for SPSU students before you present

1. Email your department coordinator asking whether an online, self-directed internship can be listed or presented, and whether it counts toward any requirement. Ask for the answer in writing.
2. Find the current Academic Regulations on the university site and look for internship duration, evaluation and document rules for your programme.
3. Ask the Career Development Cell or TPC office for the current placement rules and the Full Semester Internship details, and check whether the 2022-23 conditions still apply.
4. Ask a faculty member in Computing and Informatics to review your question and results before printing, and ask whether the department hosts poster sessions or project showcases.
5. Check the Academic Calendar 2025-26 page for dates, so your two weeks do not clash with exams or mandatory sessions.

Where to Present and How to Reuse the Poster

Start with low-stakes venues: a departmental showcase, a technical club event or a student symposium. Then convert the same material into a one-page PDF for your portfolio, a short GitHub README and a resume line. A resume line might read: “Compared three text-classification approaches on a public dataset in a 14-day online AI internship; presented findings as a poster.” Do not add claims such as published, accredited or credit-bearing unless that is documented.

Frequently Asked Questions

How many results should a poster from a 14-day ML study show?

One main results table and one chart are enough. Add a confusion matrix or failure examples if space allows. More than that usually reduces readability without strengthening the argument.

Can I present a poster if my model performed poorly?

Yes. A clearly framed negative or modest result, with error analysis and honest limits, is often more useful than a inflated one. Explain what you learned about why it underperformed.

What size should my AI research poster be?

Many events use A0 or A1, but sizes vary. Ask the organiser for dimensions, orientation and any template before you design, and set your canvas to that size from the start.

Do I need code on the poster?

No. Show a pipeline diagram and key settings instead. Add a QR code to a notes page or repository so interested visitors can see details.

Is a small public dataset acceptable for a poster?

Yes, if you state its source, licence, size and limits. Avoid generalising beyond it, and avoid datasets containing personal data you have no permission to use.

Will an SPSU department accept an online internship poster for evaluation?

The public pages reviewed do not say. Ask your department coordinator or the Career Development Cell in writing, and ask which document or regulation governs the answer.

How do I show that my result is not just random-seed luck?

Run each model with at least five seeds or cross-validation folds and report mean and standard deviation. Then compare the gap between models with that spread.

What should I put in the limitations box?

List concrete constraints: dataset size and source, single-domain data, limited tuning, no external test set and any ethical or privacy concerns. Specific limits signal rigour.

Related Artificial Intelligence guides

• Andhra Kesari University Students: Learning Python for AI in the Four Weeks Before Your Internship
• Azim Premji University Students: AI for Education Data Projects You Can Finish in an Internship
• Reva University Students: Reading Research Papers Quickly During an AI Internship
• All Artificial Intelligence internship guides