Baljit Singh Blog
Long-form writing on memory-native LLMs, symbolic RAG, neuro-symbolic systems, agentic AI, hardware-aware architecture, and the deeper structure of meaning.
How better AI answers can undermine the sources they depend on Imagine an independent reviewer who spends a month testing a new generation of batteries. She buys equipment, repeats measurements, investigates unexpected results, and publishes a careful comparison. An AI…
Vector search can help an LLM retrieve relevant text, but memory is more than similarity. Real memory needs structure, durability, revision, and trust.
The hard part of agentic AI is not making an LLM call a tool. The hard part is designing a system that can be…
RAG is easy to prototype and hard to trust. The difference is not usually the embedding model alone. It is the architecture around memory,…
When people ask what knowledge is, they often picture a mental warehouse of facts. When they ask what intelligence is, they picture raw problem-solving…