Enhancing LLM Powered Development with Clojure's REPL
summary
Clojure codebases with ~35,000 lines get translated to Kotlin by feeding targeted, repo-aware context from IntelliJ via a REPL integration, using Claude 3.5 Sonet and editor APIs to auto-generate API specs and pull only the needed files and signatures. Prompts grow large — a baseline 4,500 token artifacts prompt and file-specific prompts up to 14,000 tokens — and whole namespaces translate to working Kotlin with about 30–120 minutes of human fixup; editor chat UI produces iterated diffs, test fixes and code reviews that often succeed.
AI-written summary · about
chapters (14)
- 0:00Introduction and context
- 0:55Journey into AI tooling
- 2:50The Kotlin coroutine problem
- 4:45Code translation superpower
- 8:46Context management lessons
- 12:34Effective prompting strategies
- 14:48Building IDE integrations
- 19:28Refining LLM results
- 24:59Advanced iteration methods
- 28:49Demonstrating the tools
- 32:22Multi-file changes
- 33:45Future of AI development
- 37:41Impact on developers
- 42:29Conclusion and Q&A
