Embedded block-level LLMs (BYOM) with integrated Agentic RAG.
Play the motion
Pause
the animation
Play
(the animation is paused)
One lesson block, with its model, its field library, its gate and its record
One lesson block, drawn large. A lesson is made of blocks, and an LLM is embedded in each block: the block has a model socket, and any one of several models fits it: Claude models, through Amazon Bedrock and Anthropic's API, or your own model (BYOM). A field library is docked beneath the block. Each field has its own library: MedTech, FinTech, EdTech and LegalTech. Before it answers, the agentic RAG loop retrieves passages from the docked library, and they slide up into the block as citations, so the answer is grounded in each field's own library. Around the block runs a rim made of the authority gate, where people keep the authority, and the record. When an answer is released the rim closes: every AI act passes the one gate and lands in the record. Arrowheads show the way the data moves: the request goes down from the block to the library and the passages come back up; the model and the citations both feed the answer; and round the rim the released answer runs from the gate to the record.
Many models
Any one fits the socket
Authority gate
People keep the authority
Record
Every AI act lands here
Lesson block
An LLM in each block
Model socket
Answer
Citations
Agentic RAG
Retrieves before it answers
grounded in each field's own library
MedTech
FinTech
EdTech
LegalTech
Claude models
through Amazon Bedrock
Claude models
through Anthropic's API
Your own model (BYOM)
One lesson block, with its model, its field library, its gate and its record
One lesson block, drawn large. A lesson is made of blocks, and an LLM is embedded in each block: the block has a model socket, and any one of several models fits it: Claude models, through Amazon Bedrock and Anthropic's API, or your own model (BYOM). A field library is docked beneath the block. Each field has its own library: MedTech, FinTech, EdTech and LegalTech. Before it answers, the agentic RAG loop retrieves passages from the docked library, and they slide up into the block as citations, so the answer is grounded in each field's own library. Around the block runs a rim made of the authority gate, where people keep the authority, and the record. When an answer is released the rim closes: every AI act passes the one gate and lands in the record. Arrowheads show the way the data moves: the request goes down from the block to the library and the passages come back up; the model and the citations both feed the answer; and round the rim the released answer runs from the gate to the record.
Many models
Any one fits the socket
Authority gate
People keep the authority
Record
Every AI act lands here
Lesson block
An LLM in each block
Model socket
Answer
Citations
Agentic RAG
Retrieves before it answers
grounded in each field's own library
MedTech
FinTech
EdTech
LegalTech
Claude models
through Amazon Bedrock
Claude models
through Anthropic's API
Your own model (BYOM)