Hey PH 👋 I built I&AI Code because interview-prep & coding AI tools all want a subscription and send your code to someone's cloud. This runs a Qwen 2.5 Coder model entirely on your machine via llama.cpp: 🧩 DSA mode — write code, run it, and the AI traces your logic to find the exact buggy line 🏗 LLD mode — an AI workspace that generates class structures, writes complete code files, and patches them in place 🎤 Interview mode — 40 mock problems with full test suites and a rubric-based AI judge, scored offline No API key. No account. No telemetry. Works on a plane ✈️ The hardest engineering problem: small local models are terrible judges. Ask a 3B model to "score this answer 0–100" and you get noise. What fixed it: Deterministic bug scanners run before the LLM — classic patterns (like backtracking without undo) are caught instantly by static analysis The judge answers binary yes/no per rubric point instead of producing a score, and must quote your answer to claim a point — unverifiable claims get flipped to "no" The final score is clamped to ±30 of a deterministically computed coverage baseline It's a public beta — macOS is well-tested, Linux less so. I'm watching issues closely and shipping fixes within hours, so if something breaks on your machine, tell me and it'll likely be fixed same-day. Happy to answer anything about making small local models reliable — including everything that totally didn't work. 🙃

I&AI Code

Offline AI interview & coding coach

Offline AI interview & coding coach