Emretheus

Fire

What's in the fire right now.

Experiments live here: things I'm testing, reading and thinking about. When one matures, it leaves this page and becomes a distillation.

  1. TESTING

    Evaluation beyond vibes

    Instead of asking “does this feel better?”, I give each project a small set of test questions and a number. If the number doesn't move, nothing happened.

    When it matures → LLM evaluations, distilled

  2. EXPLORING

    Hybrid retrieval

    Word search and meaning search each catch what the other misses. I'm testing how to combine them without making the system harder to debug.

    When it matures → Reranking, distilled · related distillation ↗

  3. READING

    Agent memory that ages well

    What should an agent keep, what should it forget, and what should it just recompute? Still reading. No clean answer yet.

    No distillation planned yet

  4. THINKING

    Local inference economics

    Running a model on your own machine is slower per token than renting one. The interesting question is when that is still the right trade.

    No distillation planned yet