You are building an IDE coding assistant with two modes: inline completion (ghost text as the developer types) and a chat panel that can answer questions about the codebase and propose multi-file edits.
Inline completion is a latency problem above all else. If a suggestion takes longer than roughly 300ms it arrives after the developer has already typed past it, and acceptance collapses. Developers type continuously, so requests are fired and cancelled constantly.
The chat mode is a context problem. Repositories are large: a monorepo can be 5 million lines, and the relevant context for "why does this test fail" might be a test file, the module under test, a config file and a recent commit.
You have 500,000 developers. Source code must never leak between customers.
Build the architecture on a canvas: place the components, configure them, connect them into a data flow, and write a short reason for each one. The AI reviewer grades your design against a rubric written specifically for this problem.
Work backwards from 300ms. What does that budget allow you to run, and what does it rule out?
Why does a left-to-right model produce bad completions in the middle of a function?
A developer asks 'why does this test fail?' in a 5M-line monorepo. What context do you assemble?
Minimum 8 components · needs a wide desktop screen