Compiled Domain Expertise (CDE)
Compiling bounded AI reasoning into static, deterministic instruction sets
About
CDE proposes severing knowledge acquisition from knowledge execution. At build time, high-capacity models operate offline as an empirical knowledge compiler (Empirical Knowledge Compilation, EKC) to explore a closed domain, extract candidate reasoning structures, and surface inferential dependencies. At runtime, only surviving validated structures execute after being lowered into a Domain Instruction Set (DIS). The Quantized Semantic Bottleneck Architecture (QSBA) uses an hourglass topology that decouples continuous, high-dimensional linguistic transducers from a discrete, integer-only core, targeting exact attribution, loud syntax checks, and microsecond execution on standard CPUs within a sub-megabyte budget. The repo includes the specification, a CCPA reference instance (QSBC), an algebra domain with an ingest/review workbench (FastAPI backend + React frontend), supporting dialogues, docs, and generated slide decks.
Summary
Compiled Domain Expertise (CDE) is an architectural specification that compiles bounded AI reasoning into a static, language-agnostic Domain Instruction Set, replacing runtime neural nets with a deterministic, sub-megabyte relational graph.
Tech
- python
- fastapi
- sqlalchemy
- pymupdf
- typescript
- react
- vite
- tailwindcss
Tags
- ai
- deterministic-inference
- knowledge-compilation
- llm
- symbolic-ai