Show HN: A deterministic middleware to compress LLM prompts by 50-80% - Tech Sentiments