Reducing Agent Token Costs + RAG Beyond Semantic Search
Why agentic systems burn tokens, what retrieval misses, and how structured knowledge changes the runtime economics.
Full episode
I build the knowledge layer behind useful AI.
I work where search, data, and agents meet—building systems that make knowledge more trustworthy, and making the hard parts clear enough for people to ship.
Making knowledge usable.
Beyond Retrieval
A framework for the applicability problem: why information can be relevant, accurately cited, and still wrong for the situation at hand.
Open at PineconePinecone Marketplace
Conceived, designed, and built end to end by Roie: a platform that turns scattered, hard-won knowledge into cited applications for people and agents.
Open at PineconeVectors and Graphs: Better Together
An ontology-guided system that combines semantic similarity with explicit relationships, backed by an architecture walkthrough and live graph explorer.
Open at PineconeIdeas get sharper out loud.
Episode page
Geometry vs. Topology in AI
Full episode
Build a Customer Assistant with Multi-Domain RAG
Open at DataCamp × PineconeFrameworks for the hard parts.
Apr 2026
Pinecone Assistant: A Managed Knowledge Layer
The gap between getting a model to answer and getting an application to answer consistently, accurately, and at scale.
By Roie Schwaber-Cohen.
- Assistant
- Managed knowledge
- Production
Mar 2026
Beyond Retrieval
A framework for making applicability constraints explicit before a system retrieves and generates an answer.
By Roie Schwaber-Cohen.
- Applicability
- RAG
- Knowledge architecture
Seattle, Washington
I’ve spent nearly twenty years building software—and just as long trying to find the cleanest way to explain what the software is actually doing.
Today that means working at Pinecone on the systems and ideas that connect AI to trusted, useful knowledge.
Away from the whiteboard, I’m in Seattle making music, building small tools, and following the same instinct: understand the structure, then make it feel natural.