Build an AI Learning Plan Around a Real Project

To build a learning curriculum with AI, start with a result you want to produce, check what you can already do, and give the tool your actual weekly hours. Ask for a sequence of skills and practice tasks with a check at each step. Review the first week's work before committing to a long roadmap.

For example, "learn data analysis" is too broad to schedule. "Use SQL to produce a checked sales summary from two small practice tables" tells you what the lessons need to support. Your curriculum can then connect reading about queries to writing and checking one. A course completion percentage alone will not tell you whether that query works.

Kognivu can turn one learning goal, your starting point, and available time into a path with daily quests. It does not certify your skills, guarantee correct subject guidance, or faithfully import a course syllabus. If you have a concrete result in mind, build a learning goal in Kognivu and compare the proposed first quest with the work you can actually attempt.

Check your starting point with a small task

Write one thing you can do now without following a complete worked solution. Keep the claim narrow: "I can select columns and filter rows from one SQL table" is more useful than "I know the basics."

Then attempt a task close to the desired result. For the sales summary, select the orders from one week and explain which rows belong in the result. If that fails, your next learning block is table structure and filtering. If it works, try a grouped total. This check helps you choose a starting lesson instead of asking AI to guess your level from a label.

You may need documentation during the attempt. Record what you looked up and what you still cannot explain. Do not confuse remembering every command with understanding the task. The Odin Project's problem-solving lesson recommends understanding the problem, planning a solution, and breaking it into smaller parts; that also helps you identify what a learning plan must teach next.

Sequence the skills needed for the output

Make a short dependency list before filling the calendar:

1. What must I understand to begin the task?
2. What must I practise to produce the result?
3. How will I check whether the result is correct?
4. Which topics can wait until this version works?

For the SQL example, that might mean reading a table, selecting and filtering rows, grouping values, joining two tables, and checking the resulting totals. A dashboard framework can wait. So can learning a second database merely because an AI plan included it.

Choose one main resource and use other sources for specific gaps. The PostgreSQL tutorial introduces relational concepts and SQL, with sections on queries, joins, and aggregate functions. It is an introduction rather than a comprehensive course. If you use another database, choose its documentation and check examples against that system.

AI can propose the order, but inspect the dependencies. A task that requires joins should not appear before you have learned what the join is supposed to match. Ask the tool to explain the prerequisite or move the task.

A four-week example with two hours each week

Suppose your practice goal is a one-page sales summary. You have four 30-minute sessions each week and a working SQL practice environment. Use fictional orders and a separate product table with one row per product. Decide the questions first: total sales for a chosen week, sales by product category, and whether any orders have no matching product.

This is an illustrative curriculum, not a promise that eight hours will be enough for every beginner. If environment setup or a concept takes longer, reduce the output or extend the plan.

Week	Learn and practise	Evidence to check
1	Read the two tables; select and filter orders	The selected rows match a tiny example you checked manually
2	Group and sum the selected orders	Category or product totals agree with your hand calculation
3	Join product information to orders	Explain which rows matched and why the join did not multiply the sales total
4	Run the agreed queries and write the summary	Another person can follow the instructions and understand the checks and limits

Within each week, you might use one session to read a focused explanation, two to attempt the task, and one to check the result and choose the next step. Change that mix when the evidence calls for it. There is no universal percentage of theory, practice, and feedback that guarantees mastery.

For a tiny totals check, imagine two notebook orders worth 20 and 30 and one pen order worth 10. Your combined sales total should be 60. If a join returns 120, investigate duplicate matches before adding charts. The lesson sequence should help you resolve that concrete uncertainty.

If a concept is still unclear after the check, keep it in the next week and remove an optional feature. Do not stack a second week of dependent work on top of an untested assumption. If scheduling is the harder problem, the coding-while-working-full-time plan provides examples of allocating limited hours.

Give AI constraints and ask for a plan you can inspect

Use a prompt like this, replacing the details with your own:

My goal is a checked one-page SQL sales summary from fictional orders and product tables. I can select and filter rows, but I have not used joins. My practice environment already works. I have four 30-minute sessions a week for four weeks. Use one main learning resource. Propose only the first week in detail, plus a brief order for later skills. For each session, give a task, a prerequisite, and an observable check. Reserve time to diagnose mistakes. Explain what to reduce if a check fails. Do not add a dashboard or assume that finishing lessons proves I can do the task.

Review the answer before following it. Can you access the proposed resource? Does each session fit a plausible block? Is its check something you can perform? Does a later task depend on a skill the plan skipped? Replace invented or inaccessible resources and verify subject advice against your chosen source.

After a week, return with actual evidence: "Filtering works, but my join doubles the total. I completed three sessions, not four." Ask for the next week's plan using those facts. A revised plan should address the failed check and the available time, rather than adding more topics to compensate for a missed evening.

If you have finished a Python course and need a first independent build rather than a curriculum, use the first-project guide. It supplies a small input/output specification and checkpoints for constructing the program yourself.

Keep the next lesson connected to a result

A useful learning plan tells you what to attempt next and how to check it. Keep the output, the time budget, and the latest failed check visible. Revise the sequence when those change; do not restart the whole curriculum just to get a cleaner progress bar.

Start a project-led learning goal in Kognivu, enter your current skills and real capacity, and inspect the path and first quest. Keep your exercises and subject checks in the environment where you do the work; use the plan to decide which one deserves the next session.