CodeOak is live: Python & SQL prep that builds your roadmap, not a question bank

CodeOak is live. After months of building in private, an adaptive practice system for Python and SQL is now open to everyone — one that measures how you solve, builds a roadmap from your goal, and generates each batch of practice for the gaps actually holding you back.

The short version:

• It's not a question bank. There's no fixed pool of problems to grind. Each batch is generated on demand for your roadmap and your last batch's signals.
• AI builds the plan — it does not grade your code. Roadmaps and questions are AI-generated; correctness is decided by deterministic test cases in an isolated sandbox. Your code is never sent to an AI model.
• Progress requires proof. A half-finished batch doesn't count and doesn't unlock anything.

Why we built it

Most prep tools optimize for one number: question count. You grind hundreds of problems, feel busy, and still freeze in the interview. The problem isn't effort — it's that random volume doesn't map to your weaknesses, and a static library can't know what they are.

CodeOak takes a different stance:

• Measurement first. An assessment shows how you solve — speed, efficiency, quality, depth — before the system hands you any work.
• Direction second. That, plus your goal or job description, becomes a roadmap with a clear next step.
• Execution last. The roadmap turns into batches: seven questions, one focus, one sitting.

Practice should compound. Each batch you finish should make the next one make more sense — and the system should decide what that next one is.

What it is not: a question bank

Here's the distinction that matters most. Most platforms are one of two things: a static question bank you work through in some order, or an "adaptive" layer that simply reshuffles or re-filters that same fixed pool. CodeOak is neither.

There is no pool. When you finish a batch, the next one is generated from your roadmap topic and the performance signals from the work you just did. Two people on the same topic don't get the same seven questions — they get the seven that fit where each of them actually is. "Adaptive" here means the content itself is built for you, not that a list got sorted differently.

How grading actually works (it isn't AI)

This is the part people get wrong about AI-era prep, so let's be exact: AI does not judge whether your solution is correct. When you submit, your code runs against test cases in an isolated sandbox and passes or fails on its output — the same deterministic way a real test suite works. For the pandas and SQL tracks, you're graded on the table you produce.

AI's job is upstream of that: it turns your context into a roadmap and generates the questions in each batch. Whether your answer is right is never an AI guess — and your code is never sent to a model to find out.

# The kind of problem you'll see early on — clean, real, graded on output.
def running_total(values: list[int]) -> list[int]:
    total = 0
    out = []
    for v in values:
        total += v
        out.append(total)
    return out

What's live today

Here's what's in your hands now:

1. Two focused tracks — a Python practice track (including pandas/DataFrame challenges) and a SQL query track, both with proof-based progression.
2. An assessment and a profile — five profiles (Shark, Tiger, Eagle, Owl, Panda) that describe how you solve under pressure, so batches can be calibrated.
3. The AI Study Buddy — bring a goal, a weak area, or a job description, and it turns that into a roadmap and the next batch.
4. Canopy — a dashboard of signals over vanity metrics: accuracy, time per question, skill coverage, batch state, and what's earned next.

Where we're headed

This is a starting point, not a finish line. On the roadmap: deeper pandas coverage, more SQL patterns, and richer progress history. We'll post product updates here under the CodeOak tag — including the May roundup on signals over vanity metrics.

Want to see the engine in action on a real topic? Start with the window functions or pandas groupby explainers — then take the assessment and let the system build the rest.

FAQ

Is CodeOak a question bank like other practice sites? No. There's no fixed pool of problems. Each batch of seven questions is generated on demand from your roadmap and the signals from your previous work, so the practice fits where you actually are.

Does AI grade my code? No. Submissions run against deterministic test cases in an isolated sandbox and pass or fail on their output. AI generates your roadmap and questions; it never sees or grades your code.

What does the assessment measure? How you solve under pressure — speed, efficiency, quality, and depth — mapped to one of five profiles. It's a diagnostic, not a flattery sticker, so the batches that follow can be calibrated to you.

How does a batch unlock the next one? By proof. Each batch has a completion threshold; finish enough questions at a sufficient quality level and the next batch is generated. Fall short and the current batch stays open — more work, not a failure screen.

Is it free to start? Yes. Start with the assessment, no credit card required.

Welcome in.