Turning metacognition notes to your competitive programming coach

This is a story about a tool called cpmeta designed to help competitive programmers analyze their own thought processes while solving coding problems. Ved and the author, a friend named Prathamanvekar, started using a simple metacognition note-taking technique while working through LeetCode and Codeforces problems. They would timestamp various thoughts and observations, like reading the problem statement or getting stuck on a specific step.

After solving a problem, they would discard these notes, losing valuable insight into their cognitive blindspots and common mistakes.

The cpmeta tool aims to prevent this by analyzing these raw notes alongside the original problem statements. Using an open-weight LLM running locally on a developer's machine, cpmeta can review the notes and provide diagnostic feedback. It calculates how long the user stalled on a problem, identifies root causes of confusion, detects recurring mental traps, and even generates personalized study plans. The tool aggregates data across all sessions to show persistent weaknesses and areas of strength.

When demonstrated to his friend Ved, cpmeta was met with amazement - the friend said "Brochacho we're actually using this for our contest prep now." The demo shows cpmeta analyzing a sample problem and providing detailed insights.

Technically, cpmeta is built with a local Ollama instance running open-weight coding models. The Ollama API is accessed with strict JSON enforcement to ensure consistent coaching data. The tool runs on top of a Python Streamlit interface and persists all data in an embedded SQLite database. The whole pipeline is designed to work entirely offline, using no cloud APIs or third-party services.

By leveraging local AI, cpmeta makes deliberate practice unlimited for students and independent programmers, while also respecting privacy and firing up the brainpower of open innovation.