Learners who can train a small PyTorch model and now want to solve real vision and language problems with pretrained networks, evaluate them properly and put them behind an API.
Intermediate level
Deep Learning and Applied AI
Transfer Learning in Practice
Build practical vision and language systems with pretrained models: transfer learning for images, object detection, text classification with transformers, efficient training and a served model with tests and a latency report.
- 40hours of work
- 19parts
- Reviewedcapstone
- Lifetimeaccess
- Transfer learning
- Computer vision
- Transformers and NLP
- Hugging Face
- Model serving
- Model evaluation
At a glance
PyTorch basics (tensors, training loop, DataLoader, evaluation), Python and basic web API ideas. A free cloud notebook with a GPU is recommended, but every exercise can also run on a CPU with smaller data. If you have finished Deep Learning Foundations: Build Neural Networks with PyTorch, you are ready.
Pass the quizzes and labs, then submit a capstone project. The NIBVEDA team reviews it and either accepts it or sends it back with comments you can act on.
What you will be able to do
- 01Adapt pretrained image models to new tasks with feature extraction and fine-tuning
- 02Run and evaluate object detectors using IoU and average precision ideas
- 03Fine-tune a transformer for text classification and compare it with a simple baseline
- 04Train efficiently with mixed precision, gradient accumulation and good checkpoints
- 05Serve a model through an API and measure its latency, robustness and fairness
Curriculum 40 hours in 19 parts
Select a kind to highlight its parts.
Stage 1A transfer-learning classifier for your own data7 parts · 11 h
- 1From toy models to real systemsLesson · Short, focused reading with questions to check yourself.1 hFree
- 2Transfer learning for imagesLesson · Short, focused reading with questions to check yourself.1 h
- 3Fine-tune a pretrained ResNetWorkshop · Follow along and build, with the steps and the traps written out.2 h
- 4Object detection, segmentation and how they are measuredLesson · Short, focused reading with questions to check yourself.1 h
- 5Run, measure and use an object detectorWorkshop · Follow along and build, with the steps and the traps written out.2 h
- 6Check yourself: transfer learning and detectionQuiz · Twelve questions, marked at once, with the reason behind each answer.1 h
- 7A transfer-learning classifier for your own dataLab · Hands-on work. Many labs are checked automatically against what must exist.3 h
Stage 2A fine-tuned text classifier with a baseline and error analysis6 parts · 10 h
- 8Text as data: tokens, embeddings and attentionLesson · Short, focused reading with questions to check yourself.1 h
- 9A text baseline, embeddings and attention by handWorkshop · Follow along and build, with the steps and the traps written out.2 h
- 10Fine-tuning pretrained language modelsLesson · Short, focused reading with questions to check yourself.1 h
- 11Fine-tune a transformer and compare with the baselineWorkshop · Follow along and build, with the steps and the traps written out.2 h
- 12Check yourself: tokens, attention and fine-tuning transformersQuiz · Twelve questions, marked at once, with the reason behind each answer.1 h
- 13A fine-tuned text classifier with a baseline and error analysisLab · Hands-on work. Many labs are checked automatically against what must exist.3 h
Stage 3A tested, containerised model API with a latency report4 parts · 8 h
- 14Train efficiently and reproduciblyWorkshop · Follow along and build, with the steps and the traps written out.2 h
- 15Serve a model through an APIWorkshop · Follow along and build, with the steps and the traps written out.2 h
- 16Check yourself: efficient training and model servingQuiz · Twelve questions, marked at once, with the reason behind each answer.1 h
- 17A tested, containerised model API with a latency reportLab · Hands-on work. Many labs are checked automatically against what must exist.3 h
Stage 4Test robustness, slices and calibration1 part · 3 h
- 18Test robustness, slices and calibrationLab · Hands-on work. Many labs are checked automatically against what must exist.3 h
FinishCapstone project1 part · 8 h
- 19A pretrained-model system, from data to a served and evaluated productCapstone · A full project for your portfolio.8 h
What you take away
An internship certificate with your skills on it.
Made for you automatically when you finish the last part, it names the internship programme, the level and the 40 hours, lists what you showed, and carries a number and a QR code. Anyone can check it on the NIBVEDA verification page, and it shows as withdrawn there if it is ever revoked.
- Proof you completed all 40 hours of this internship programme
- An Internship badge for your profile, CV or portfolio
- Not a job, a stipend or an academic degree, and it says so
Keep building.
3 programmes in this track, from beginner to expert. Buy the learning path and save 15%.
Learn how neural networks learn by building them yourself in PyTorch: tensors, gradients, training loops, data pipelines, evaluation and convolutional networks, finishing with an image classifier you can explain and defend.
- PyTorch
- Neural networks
- Gradient descent
Engineer reliable applications on top of large language models: prompting and structured output, retrieval-augmented generation, evaluation harnesses, tool-using agents, security and red-teaming, and cost, latency and quality trade-offs in production.
- LLM application design
- Retrieval-augmented generation
- Evaluation harnesses
See all 3 programmes in Deep Learning and Applied AI Take the placement test
Quick answers.
The policies have the full detail.
Is this a real internship?
It is a virtual internship programme: a structured, self-paced course built around a real project, with checked assessments and a certificate. It is not employment, it does not carry a stipend and it is not an academic or AICTE-approved qualification. What it gives you is verified, project-based evidence of skills.
How does self-paced work?
You start when you like and move at your own speed. Each part unlocks when you finish the one before it. There are no live classes and no deadlines. Your access is shown on each programme page.
How is my work checked?
Quizzes are marked as soon as you submit them. Labs use a tick-list of tasks. Intermediate and Expert programmes finish with a capstone that the NIBVEDA team reviews and either accepts or sends back with comments.
Can I get a refund?
Please read the refund policy before you buy. It explains when a refund is possible.
Ready to start?
Rs. 824.82 including GST. Lifetime access. Read the first lesson free before you decide.