How to roll your own LLM
summary
Claim of 4,500 developer years saved by Amazon for Java upgrades anchors the scale question, while concrete tooling includes OpenRewrite recipes piped to LLMs, ADA-style repo agents, Olama for local models, Hugging Face Text Generation Interface/VLM on H100 80GB nodes and AMD MI325X clusters. Migration work uses a custom parser + knowledge graph for legacy PLI/COBOL and generates Playwright tests; experiments include a fine-tuned OCR that preserves headers and merged models (R1+v3) that trade reasoning quality for speed.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:20Why should we care
- 1:40AI assisted coding
- 5:28Code migration
- 15:12Recap
- 18:20Infrastructure
- 22:24Experimentation
related talks
26mReverse Engineering Using LLMs
01h 49mAI engineering with open access LLMs that lie, curse, and steal
50mHow to Test LLM Agents
29mThe Evolution of LLMs: From Imitation to Reasoning
30mLLMs: Beyond the Hype - A Practical Journey to Scale
45mEverything You Need to Know About Running LLMs Locally
25mvLLM: Easy, Fast, and Cheap LLM Serving for Everyone
29mThe Evolution of LLMs: From Imitation to Reasoning