You have reached the last lesson. In 10-01, 10-02 and 10-03 you gathered books, courses and communities; what remains is to decide where to go with all of it. This lesson is a compass: first an honest self-assessment of what you have learned in the nine previous modules (with a concrete way of checking each competence), then the map of AI professional profiles (what each one does, which competences it demands, which modules of the course are its base and what its next step is), and after that four detailed learning paths (data engineering and classical ML; deep learning and NLP/LLMs; knowledge-based systems and probabilistic reasoning; AI project management and governance), each with stages, resources from 10-01 and 10-02, a portfolio project and a checkpoint milestone. We add a guidance note on certifications and formal education, how to build a portfolio that speaks for you and how to stay current without being overwhelmed. As an example of two different paths, the next steps of Marta and Diego at NovaMarket and the company's twelve-month roadmap; and for you, a 90-day personal plan template. We end with the recap of the ten modules and a farewell.

1. Self-assessment: what you have learned and how to check it
2. Map of AI professional profiles
3. Path A: data engineering and classical ML
4. Path B: deep learning and NLP/LLMs
5. Path C: knowledge-based systems and probabilistic reasoning
6. Path D: AI project management and governance
7. Certifications and formal education, with caution
8. Portfolio and how to stay current without being overwhelmed
9. Next steps for Marta and Diego and NovaMarket's 12-month roadmap
10. 90-day personal plan: template
11. Common Mistakes and Tips
12. Exercises
13. Course recap and farewell

Before choosing a path, measure where you stand. The check must be something you can do, not something that "rings a bell".

Score each row from 1 (I could not do it) to 3 (I could do it fluently). The rows with a 1 are your reinforcement priorities, whichever path you choose.

Job titles vary between companies and countries, and in small ones a single person covers several. What matters is understanding the content of each profile.

A junior does not have to choose for ever: the profiles share a great deal and switching is normal. It is worth choosing a clear first step, though; that is what the paths are for.

For those who enjoyed module 4 and the returns predictor, and want to be a data scientist, ML engineer or MLOps engineer.

Portfolio project: a repository with a complete case (definition, data, pipeline, evaluation, API or batch process, model card, drift monitoring) on public data, with a README that explains the business decision. Checkpoint milestone: someone else clones it and runs it with a single command, and you can explain every choice in a thirty-minute technical interview.

For those who clicked with module 5 and the reviews case, and want to be an NLP or vision specialist, or an AI engineer focused on LLMs.

Portfolio project: a RAG assistant or a fine-tuned text classifier, with your own evaluation set, error analysis (negations, irony, out-of-domain cases) and an honest model card. Checkpoint milestone: explain with a diagram what attention does and why RAG reduces (but does not eliminate) hallucinations, and show an evaluation table with cases where your system fails.

For those who found their way of thinking in module 6 and in the capstone project (rules + Bayesian network + classifier): knowledge engineering, decision systems, regulated domains where explainability rules.

Portfolio project: an explainable decision system (rules + probability + a model) with documentation of every rule, its origin and its justification, and an interface that shows "why" for each decision. Checkpoint milestone: interview an expert in some domain (even a colleague), formalise ten of their rules and have them recognise their knowledge in the system's output.

For those who connect with module 8, the ethics of 02-04 and Diego's role: defining use cases, measuring value, managing risk and compliance, coordinating technical teams without necessarily being the one who programs.

Portfolio project: an AI roadmap for a real or fictional organisation (NovaMarket will do) with prioritised cases, risk analysis under the AI Act, a governance plan (who approves, who audits, what documentation) and indicators. Checkpoint milestone: present it in twenty minutes to someone technical and someone from the business side, and have both understand it and challenge it sensibly.

• MOOC certificates (Coursera, edX, DeepLearning.AI): they attest that you completed the course; they add to a junior CV, especially alongside the corresponding project in your portfolio. They do not replace demonstrable practice.
• Cloud-provider certifications (the ML/AI ones from the major providers): valued when the job is tied to that platform; their syllabus changes frequently and it is best to prepare them once you are working, or about to work, with that cloud.
• Formal education (degrees, university master's in AI, data science or similar, vocational programmes with an AI and big data specialisation): they provide a mathematical base, depth and a network; they are a big investment of time and money that makes sense if you want research, very technical profiles or a deep career change. No qualification guarantees a job on its own; what opens doors is the combination of a solid base, visible projects and the ability to explain what you do.

Rule of thumb: portfolio and practice first; certificate as a complement; formal education as a considered decision according to goal and circumstances.

Portfolio. One public repository (or two) that anyone can understand in five minutes:

• A clear structure like novamarket_ai/ (07-04): a README with the problem, the decision it supports, the data (public or synthetic, never confidential), how to run it with one command, results with the metric that matters and limitations.
• Model cards for every model (08-01): intended use, data, metrics by subgroup, risks. That is what distinguishes a professional portfolio from a collection of notebooks.
• Reproducible notebooks: fixed seeds, declared dependencies, saved outputs, no local paths or keys.
• Two or three well-finished projects are worth more than ten half-done. One per chosen path is enough to start.
• Write a short post (blog, long README or a thread in a community from 10-03) recounting what you learned; communicating is part of the competence.

Staying current without being overwhelmed. The pace of the sector is impossible to follow in full, and there is no need to:

• One weekly newsletter (The Batch, for example) and one community; everything else, muted.
• One paper a month with the three passes from 10-01; one new tool a quarter, tried out in a notebook.
• Distinguish fundamentals (change little: everything in modules 3 to 6) from tools (change a lot: the LLMs of the moment, the agent libraries). Invest quality time in the former; give the latter rationed time and curiosity, without anxiety.
• Review your plan every 90 days (section 10) and adjust it; do not do it every week according to the latest news.

Two people, two different paths, one company.

Marta (head of data; paths A and B):

1. MLOps and monitoring of the returns predictor. Automate batch retraining with versioning (MLflow), drift alerts like those from 09-04, and a quarterly review of the model card with metrics by subgroup (country, amount band). Resources: Huyen, MLflow and FastAPI documentation, an MLOps course.
2. Build the case 7 RAG assistant, so far only designed: an internal pilot over the returns policy, an evaluation set of 100 questions, explicit limits ("it does not answer about specific orders"), risk review under the AI Act. Resources: Hugging Face Course, RAG short courses, 05-05 reread.
3. Specialise the team: the Machine Learning Specialization as a common base, one member towards NLP (path B) and another towards data/MLOps (path A); a communities rule with no real data (10-03); one internal portfolio project per person per half-year.

Diego (operations; path D):

1. AI literacy for operations: Elements of AI and the Crash Course without the code; a fixed list of questions for every model presented to him (intended use, data, business metric, subgroups, drift, who supervises).
2. Governance and the AI Act: classify NovaMarket's nine use cases by risk together with Marta; define the human oversight of the predictor's queues (who reviews "review", within what deadline) and the minimum documentation per case.
3. Define new use cases with the one-page document from 08-01: staffing forecasts at peaks, real-time incident prioritisation, and decide which ones not to do (for risk or for low value).

NovaMarket's 12-month roadmap (an example, not a recipe):

Fill it in yourself, with your path and your real time. It is the same logic as the path stages, compressed into one quarter.

• Choosing a path by fashion (all LLMs) rather than by what you enjoyed and what your local market asks for. All four paths lead somewhere; the one you will finish is the one you like.
• Planning twelve months in detail. Plan 90 days in detail and the rest in broad strokes; review every quarter.
• A portfolio of notebooks without context. Without a README, without a model card, without a business decision, a notebook says nothing about you.
• Waiting to "be ready" to take part or apply. A junior with a well-documented project and the ability to explain it is already ready to start.
• Confusing certificate with competence or qualification with guarantee: they complement, they do not replace.
• Anxiety about the latest news. Fundamentals with quality time; tools with rationed time. What is indispensable today will be called something else in two years; what you learned in modules 3 to 6 will still hold.
• Tip: write your plan, put dates on it, tell someone about it (a colleague, a community, your own Marta or Diego) and review it after 90 days. Consistency beats talent among self-learners.

1. Self-assessment and priority. Fill in the section 1 table with your scores (1-3), identify the two lowest rows and write down what you will do in the next two weeks to raise them by one point, with a specific resource and practice.
2. Choose a path and a project. Choose one of the four paths (or a combination of two), justify in a paragraph why (what you enjoyed, which profile attracts you, what your environment asks for) and define your portfolio project: problem, data (public or synthetic), deliverables and checkpoint milestone.
3. 90-day plan. Fill in the section 10 template completely for your case, with real dates.

1. Example: "Lowest rows: module 5 (backpropagation, 1) and module 6 (Bayesian networks, 1). Two weeks: watch the four 3Blue1Brown videos and redo the two-layer network from 05-03 in NumPy without looking (week 1); reread 06-03 and program inference by enumeration on the incidents network with a new case (week 2). Check: explain both to a colleague in ten minutes."
2. Example: "Path A with a touch of MLOps: I enjoyed the returns predictor and the batch deployment; the ML engineer profile fits my programming background; in my city there is demand for data in production. Project: demand forecasting with a public retail-sales dataset; deliverables: a repository with a scikit-learn pipeline, temporal validation, a FastAPI API, a model card, a drift check; milestone: a colleague clones it, runs make demo and gets the predictions and the model card in under five minutes."
3. Example (abridged): Wks 1-2: StatQuest on cross-validation and Géron ch. 2; returns predictor notebook with nested validation; subscription to The Batch and a Stack Overflow account. Wks 3-6: ML Specialization course 1; Kaggle Titanic; one question with an MRE if I get stuck. Wks 7-10: demand project with API and model card; publish and ask for a review on r/learnmachinelearning. Wks 11-12: three-pass reading of the Model Cards paper; blog post; new plan.

Before we say goodbye, the full view of the ground covered, with the NovaMarket thread that has accompanied us since the first module:

You started this course with some Python and curiosity; you finish it knowing what AI is and what it is not, how search works, how to learn from data, how a network learns, how to reason with rules and with probabilities, which tools it is built with, how to take a project from the idea to a documented prototype and, since this module, how to keep learning with judgement and in good company. None of this makes you an expert yet, and it does not need to: it makes you someone who knows where to start and how to check they are making progress, which is exactly what is asked of a serious junior professional.

Marta and Diego carry on at NovaMarket with their roadmap; you have yours in section 10. Choose a path, write your 90-day plan, build a project you can explain with pride, ask well when you get stuck and help whoever comes after you. AI is a huge, changing field, but its fundamentals are stable and now they are yours. Thank you for coming this far. Here ends the Fundamentals of Artificial Intelligence course. Good luck, and enjoy the road ahead.