Reverse Engineering Using LLMs
summary
Claims that LLMs convert COBOL to Python/Java quickly using a pipeline of tokenization, semantic chunking, fine-tuning and CI/CD, with CodeLlama and Hugging Face Transformers as core tools. Reported results: one bank achieves ~60% faster and ~40% cheaper rewrites, another converted 5 billion lines in 9 months while maintaining functional parity. Built artifacts include automated translators plus an “LLM-as-judge” stage and human-review loop; failures are hallucinated logic and IP leakage, mitigated by human validation, automated tests and governance.
AI-written summary · about
chapters (11)
- 0:00Reverse engineering with LLMs
- 1:49Legacy systems modernization delay
- 4:18Legacy code migration challenges
- 5:25LLMs understand legacy code
- 8:18AI and human expertise hybrid
- 12:39LLM code conversion workflow
- 14:27LLM modernization case studies
- 16:35LLMs vs manual migration
- 17:20LLM adoption risks and mitigation
- 18:24AI landscape 2025 trends
- 19:29Q&A
related talks
29mHow to roll your own LLM
47mValidating LLMs – A True Story
15mFrom Words to Data: Using LLMs to Turn Questions into SQL
01h 49mAI engineering with open access LLMs that lie, curse, and steal
51mPractical LLM Inference in Modern Java
29mLeveraging open-source LLMs for production
47mHarnessing LLMs for Next-Level Security Analysis
20mKnowledge Transfer from High-Resource to Low-Resource Programming Languages for Code LLMs