CodeOak is adaptive AI programming prep for Python, SQL, and pandas — built for both skill-building and interview readiness. The system evolves from how you actually solve problems, not how much content you consume.
Python + SQL
Focused tracks, deeper product behavior
Assessment
Measurement layer
AI roadmap
Direction layer
What CodeOak is
CodeOak is built around a simple idea: measurement, direction, and execution should not be mixed together. Assessment measures. AI gives direction. Batches turn that into real work.
That separation keeps the product honest. It also keeps progression adaptive. The system moves forward based on what your work shows, not on how much content you clicked through.
CodeOak is
CodeOak is not
Why It Was Built
AI has fundamentally changed how people learn, upskill, and get hired — but most coding platforms have not kept up. The old “solve random questions” model made sense when grinding volume was the only way to prepare. In an era where AI can generate targeted practice on demand, that approach is no longer enough. What matters now is structured, goal-driven preparation — not just more problems.
That gap showed up firsthand. Using AI to generate questions and building a custom learning flow manually worked — but it was time-consuming and hard to sustain. Figuring out what to practice next, whether progress was heading in the right direction, and how to stay focused on a specific goal all required constant manual effort that the tooling should have handled.
CodeOak was built to close that gap — an adaptive system designed for this moment, where AI does the heavy lifting of turning your goal into a structured roadmap with the right questions, clear progression, and interview-focused practice built in from the start.
Who It Is For
Developers building real Python and SQL skills through deliberate practice.
Candidates preparing for Python and SQL interviews with structure, not volume.
People who want to know exactly where they stand and what to do next.
What Makes It Different
Roadmaps are built from your context — a goal, a job description, a weak area — not from a generic curriculum written before you arrived. Each batch is then generated on demand, shaped by the performance signals from the work you already did.
Progress requires proof. A half-finished batch does not count and does not unlock anything. When performance stalls, the system makes it visible and changes what comes next — not pretends it did not happen.
Product Questions
These are the questions people usually ask when they want to understand what is live today and how the system is meant to work.
Q
Assessment is the measurement layer. It shows how you solve under pressure. The roadmap is the direction layer created from your goals and context. Batches are the execution layer where that plan turns into real Python or SQL practice.
Q
AI turns your goal, weak areas, or job-description-style context into roadmap direction. It decides what to generate next and how the system should continue. It is there to guide the work — not to be an endless generic chat layer.
Q
No. Code evaluation is deterministic — your submission runs against test cases and passes or fails based on the output. AI models never see or grade your code.
Q
Each batch has a completion threshold. When you finish enough questions at a sufficient quality level, the next batch can be generated. If the threshold isn’t met, the current batch stays open — you work more on it, not a different one.
How AI is Used
CodeOak uses AI in two specific places: building your roadmap from context, and generating the questions in each batch. Outside of those two roles, AI does not touch the product.
Roadmap generation
When you provide a job description or learning goal, that context is sent to an LLM — primarily Gemini, with Groq Llama and Qwen models as fallback — which returns a structured roadmap with skill priorities, levels, and a first batch to start from.
Question generation
Each batch of seven questions is generated on demand by Groq (Llama 4 and Qwen models, Gemini as fallback), shaped by the roadmap topic and your performance signals from previous batches. Questions are built to your current point in progression — not drawn from a static pool.
What AI does not do
AI does not evaluate your code. Submissions run against deterministic test cases in an isolated sandbox. AI shapes what you practice and what comes next — whether your solution is correct is not an AI judgment call.
Your Data
No vague privacy language. Here is exactly what happens to the three main types of data in CodeOak.
Identity & auth
Authentication is handled by Supabase — an independent auth provider. CodeOak never stores your password. Sessions use short-lived JWTs in HttpOnly cookies. Google and GitHub OAuth sign-in are supported.
Code & submissions
Your code is saved against your account so sessions can resume where you left off. Execution happens in isolated Docker containers with no network access — memory-limited, time-limited, and sandboxed. Your code is never sent to any AI model.
AI context
When you use the AI Study Buddy, your job description or learning context is sent to Gemini or Groq to generate your roadmap and batch questions. This context is stored in your account and is not shared with other users or used to train models.
Founder / Team
CodeOak stays centered on Python and SQL because depth matters more than pretending to support everything equally well.
The next step changes when performance changes. Completion thresholds and batch signals matter more than passive completion.
AI in CodeOak has a specific job: build the roadmap, shape the next batch. It is not there to simulate conversation or pad the feature list.
Start with the assessment. Then let context-driven roadmap work and adaptive batches carry the rest of the system forward.
Free to start · No credit card required
