Case study
CodePop
A solo AI code intelligence platform: real-time code explanation, debugging assistance, and algorithm generation, built on OpenAI tool calling with context-aware prompts.
- Your code: The user pastes code, an error or an algorithm problem.
- Prompt design: Builds a prompt that asks for a walkthrough rather than a generic chat reply.
- OpenAI API: Tool calling produces structured output for each flow.
- Answer: An explanation, a diagnosis with a fix, or a generated solution with approach notes.
Context
Engineers spend a lot of time on code they did not write. CodePop compresses that loop: paste a snippet, get an explanation, a fix, or a generated solution.
Problem
Raw LLM output often sounds like a chatbot. The product needed prompts and UX that feel closer to a careful walkthrough than a generic paragraph.
My role
Solo builder: frontend, API, prompt design, and deployment.
Challenge
Getting specific output from a general model. Context has to be enough, and the interface has to keep the user in a tight loop rather than a chat dump.
Approach
Three flows: structured explanations, algorithm solutions with approach notes, and debugging that explains the failure instead of only patching it. Prompting is the product as much as the UI.
Outcome
Live at codepop-one.vercel.app. A contained AI product I designed and shipped end to end.
Decisions
Interface over model-chasing
The lesson was that prompt design and a calm UI matter more than swapping models. The constraint was usefulness, not novelty.
What shipped
Contextual explanations
Intent first, then detail where it matters — closer to how a senior engineer would walk through unfamiliar code.
Debug with a reason
Diagnosis plus a corrected version, with an explanation of why the bug happened.
Stack
Next.js · React · TypeScript · Node.js · Express · OpenAI API · Vercel
A small AI product, built to see whether careful prompting and a tight UI beat a generic chat box.