SCOI Atlas · L2 Models
L2C Embedding & Retrieval
Text, image and code embeddings; rerankers; hybrid search.
TL;DR · Direct answer
L2C Embedding & Retrieval is a sublayer of L2 Models in the Supply Chain of Intelligence™ (SCOI) by Anand Arivukkarasu. It is a required capability but not on its own a durable moat. Treat embeddings as commodity infra. Spend your effort on the retrieval graph and reranking, not on evaluating five embedding vendors.
What actually matters at L2C
- Reranker quality contributes more to RAG accuracy than embedding model choice.
- Hybrid (BM25 + dense) beats pure dense in most enterprise settings.
- Embedding models are cheap; the moat is the corpus you embed.
The startup lens
Treat embeddings as commodity infra. Spend your effort on the retrieval graph and reranking, not on evaluating five embedding vendors.
Vertical lens — how this plays across categories
Enterprise search
Chunking + graph construction is where differentiation actually lives.
Legal / medical
Domain-tuned rerankers on proprietary judgement data.
Developer tools
AST-aware chunking beats semantic chunking for code.
How to defend L2C
- Proprietary reranker trained on customer outcome data (L1D linkage).
- Index infrastructure that stays consistent across model swaps.
Other sublayers in L2 Models
L2A
Foundation & Multimodal Models
Frontier models from OpenAI, Anthropic, Google, Meta, xAI, Mistral, DeepSeek and their open-weight peers.
L2B
Specialized & Fine-Tuned Models
Domain-tuned models — code, biology, legal, finance, radiology, robotics.
L2D
Model Routing & Composition
Cost-quality routers that pick the right model per request; mixture-of-experts style composition.
L2E
Reasoning & World Models
Chain-of-thought, tree search, o-series style reasoning, and true world models for physical AI.