Roadmap · Specialise
Machine learning (Computer vision)
Opens with creditsFor programmers and students who want to train models on data rather than only call someone else's API. By the end you can clean a real dataset, choose and honestly evaluate a classical or deep learning model, and ship it as a reproducible, monitored service.
Curated by Tauseef Fayyaz8 stages, 72 topics~23 months at a few hours a weekBy the end, you will be able to
- Load, clean and explore a messy real dataset in pandas and explain what you found with charts
- Train, tune and compare scikit-learn models with proper validation and no data leakage
- Pick the right metric for a problem and explain bias, variance and leakage using your own projects
- Build, train and fine-tune neural networks in PyTorch for images or text
- Serve a model behind an API with tracked experiments, a versioned model and drift monitoring
The path
8 stages, in the order to learn them. Open one to start.
1Python for dataThe language and the libraries every ML codebase leans on: core Python first, then NumPy arrays, pandas DataFrames and plotting, working in Jupyter notebooks.9 topics · ~3 months
- 1Core Python for data workBefore any library, plain Python needs to feel automatic.
- 2Working in Jupyter notebooksA Jupyter notebook is one document that mixes code, its output, charts and your notes.
- 3NumPy arrays, shapes and indexingA NumPy array is a grid of numbers that all share one type, stored side by side in memory so the computer can process them quickly.
- 4Vectorisation and broadcastingVectorisation means working on a whole array in one go, such as
prices * quantities, instead of looping over items one by one in Python. - 5pandas DataFrames: selecting, filtering and cleaningA pandas DataFrame is a table.
- 6groupby, merge and reshapingMost analysis questions follow one pattern: split the rows into groups, work out something for each group, then combine the answers.
- 7Charts that make one clear pointCharts are how you spot problems in data and how you explain results to people who will not read a table.
- 8CSV, JSON and ParquetData arrives in a few formats, and each has quirks.
- 9Images as NumPy arraysTo a computer, a colour photo is just a NumPy array of shape (height, width, 3): one number from 0 to 255 for the red, green and blue channel of every pixel.
9 in this stageOpen this stage2The maths you needThe linear algebra, calculus, probability and statistics that show up in every model and every evaluation. Aimed at intuition and working in code, not proofs.9 topics · ~3 months
- 1Vectors, matrices and the dot productIn ML, one data point is a vector, which is just a list of numbers.
- 2Matrix multiplication and shapesYou can multiply a matrix of shape (m, n) by one of shape (n, p), and the answer has shape (m, p).
- 3Derivatives and gradientsA derivative tells you how fast a function's output changes when you nudge its input a little.
- 4The chain ruleThe chain rule gives the derivative of a function inside another function.
- 5Gradient descentGradient descent is how almost every model learns.
- 6Probability and Bayes' theoremProbability is the language of uncertainty, and it runs through ML.
- 7Probability distributionsA probability distribution describes how likely each possible value is.
- 8Descriptive statisticsBefore modelling anything, summarise it.
- 9Hypothesis tests, p-values and confidence intervalsA hypothesis test asks whether a difference you saw could just be noise.
9 in this stageOpen this stage- Data cleaning and feature engineeringWhere most of the real work in ML happens: handling missing values and outliers, exploratory analysis, encoding categories, scaling, and building features that actually carry signal.9 topics · ~7 weeksOpens with credits
- Classical machine learningThe algorithms that still solve most tabular business problems: regression, classification, trees and gradient boosting, clustering and PCA, with scikit-learn as the working tool.10 topics · ~4 monthsOpens with credits
- Evaluating and tuning modelsKnowing whether a model is actually good: metrics that match the problem, cross-validation, avoiding leakage, hyperparameter search, and explaining what the model relies on.9 topics · ~6 weeksOpens with credits
- Deep learningNeural networks from backpropagation up to transformers, trained in PyTorch, with CNNs and transfer learning on images as the thread to follow. The next stage turns this into detection, segmentation and vision transformers.10 topics · ~4 monthsOpens with credits
- Computer vision in practiceFrom classification to the tasks real vision products need: OpenCV for classic image work, labelled datasets, augmentation, object detection, segmentation, vision transformers, the metrics that judge them and running a model on a device.8 topics · ~3 monthsOpens with credits
- Production ML and MLOpsTurning a notebook model into a system that keeps working: reproducible training, experiment tracking, a model registry, serving behind an API, CI for models, and drift monitoring with a plan to retrain.8 topics · ~3 monthsOpens with credits
6 more stages, 54 more topics, free with an account.
You have seen the open stages of the Machine learning roadmap. Every topic it covers is listed above, and an account opens the rest: what each topic means, a question to check yourself, the courses, the practice and Tauseef's advice for every stage. Early access is open: sign up and your account is active shortly after.
Everything here is free to learn from. A few things, like adding a roadmap, keeping a cheat sheet or asking Tauseef a question, use credits. You get 15 when you join, your first roadmap is free, and if you are learning seriously and run out, you can ask for more. Credits only make sure the platform goes to people who are actually learning.
Build a project
A finished project proves more than a certificate.
- Start with this onebeginner1 semester · 2 studentsFabric Defect Inspection with Anomaly Detection for Small Textile Units
A beginner-sized vision project where defects are rare, so augmentation and metrics matter
- Python
- PyTorch
- Anomalib
- OpenCV
- FastAPI
- intermediate2 semesters · 2 to 3 studentsConstruction Site PPE Compliance Monitoring with Privacy-First Video
Object detection on real site video, judged by per-class recall on small, partly hidden objects
- Python
- Ultralytics YOLO
- ByteTrack
- OpenCV
- FastAPI
- advanced2 semesters · 3 to 4 studentsShelf Gap and Planogram Check from Phone Photos for FMCG Distributors
Detection on crowded phone photos with many near-identical classes, then a report people act on
- Python
- PyTorch
- Ultralytics YOLO
- DINOv2 or CLIP embeddings
- FAISS
Related on Sefism
Picked from the other libraries by shared topics.
Best practices
See allChecklists
See allProject structures
See all- Machine learning project project structureFor students and analysts whose model works in a notebook and now needs to be reproducible, reviewable, served and watched. It starts from the Cookiecutter Data Science v2 layout (immutable raw data, numbered notebooks, a real Python package) and grows into versioned data with DVC, typed parameters, data validation with pandera, experiment tracking and a model registry with MLflow, a FastAPI prediction service, drift monitoring and, for a team, scheduled retraining, model cards, fairness checks and shadow deployment.
- MLOps pipeline project structureFor engineers who can train a model and now need the training to be a pipeline anyone can rerun, review and release. A uv workspace holds the pipeline package (prepare, validate, featurize, train, evaluate, register steps run by DVC), a small shared feature package, and a separate FastAPI prediction service that loads models from the MLflow registry by alias. It grows into CI that reproduces the pipeline on pull requests, scheduled retraining, approved promotion and monitoring.
Career questions
See all- My machine learning projects never leave a Jupyter notebook. How do I learn to deploy a model?Move the notebook code into a small Python package, save the whole preprocessing and model pipeline, put it behind a FastAPI endpoint in a Docker container and deploy it. Then add what makes it production: tracked runs, a registry, a test in CI, prediction logging and a drift check. Start with batch predictions if you can.
- Should I learn AI and machine learning, or become a strong software engineer first?For most people, become a solid software engineer first, then add AI on top. Most AI jobs today are engineering jobs: building products around models, data pipelines and APIs. Research-heavy machine learning is a different path that needs strong maths and usually a postgraduate degree.
- How do I move into AI engineering, data science or DevOps?Pick the field by the work you want to do every day, not by the hype. Use what you already know as a bridge: developers usually reach AI engineering or DevOps fastest, analysts reach data science fastest, and a portfolio of one or two serious projects matters more than certificates.
Cheat sheets
See all- MLOps and LLMOps cheat sheetShip, serve and monitor models.
- PyTorch cheat sheetTensors, autograd, nn.Module, training loops, data loading, GPUs, mixed precision, torch.compile, DDP and export in PyTorch.
- Hugging Face cheat sheetThe Hub, transformers, tokenisers, datasets, fine-tuning, LoRA, quantisation, inference and embeddings with the Hugging Face libraries.
Work with me
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Sessions are free for approved Sefism members, and a few slots open each week.
- 1:1 Mentorship
- Fresh Graduate Mentorship
- Career Guidance
Brand deals, promos and partnerships: tauseeffayyaz@gmail.com