I'm working on a school project about a deterministic code translation tool, but before I start building it, I want to understand whether this is still a meaningful problem or whether existing translators and AI have mostly solved it. Have you ever needed to convert code from one programming language to another? What approach did you use—AI, an automated translator, or manual rewriting—and how reliable was the result? I'm especially interested in cases where the generated code was incorrect, inefficient, or subtly changed the original behavior. Honest criticism would help me identify the limitations of current tools and decide whether there is a worthwhile problem for my project to address.
4 Answers
The general problem is usually called transpiling, compiling, or porting, depending on what is being converted. Tools such as the TypeScript compiler and CoffeeScript handle common language-to-language transformations, while porting can also involve moving code to a different platform. AI is another option now, but it still needs careful review.
The assignment may be primarily about learning parsing, language structure, and translation—not necessarily creating a tool that solves an untouched commercial problem. Even if AI can generate a translation, someone still needs enough understanding to verify whether the result is correct. Building a small translator can be valuable for that reason.
There are definitely real use cases. For example, Java-to-JavaScript tools can make existing code usable on another platform. The difficult part isn’t just producing syntax that runs—it’s preserving the original behavior. A deterministic translator can repeat the same mistake every time if it doesn’t account for differences in operators, types, libraries, edge cases, and performance.
A good school-project scope would be one specific language pair with a deliberately limited feature set. Clearly reject unsupported constructs instead of generating questionable output, then compare your results with an existing translator and an AI system using the same test suite. That would let you demonstrate exactly where each approach succeeds or fails.
That makes sense. I’m not planning to have AI write the project for me—I’m trying to understand whether the problem itself is already well covered and where current tools still struggle.