# US 2026/0093679 A1 · Systems and methods for an indexing tree based artificial intelligence conversation agent

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0093679 A1
- Document ID: US20260093679A1
- Status: Published application
- Kind: A1
- Application number: 19/041,855
- Filed: 2025-01-30
- Published: 2026-04-02
- Earliest claimed: 2024-10-01
- Assignees: Salesforce, Inc.
- Applicants: Salesforce, Inc.
- Inventors: Nan Zhang; Prafulla Kumar Choubey; Alexander R. Fabbri; Gabriel Bernadett-Shapiro; Chien-Sheng Wu
- Web page: https://patentgrep.com/patent/US20260093679A1

## CPC

- G06F 16/2246: {Trees, e.g. B+trees}
- G06F 40/35: Discourse or dialogue representation
- G06N 3/0475: Generative networks
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Embodiments provide a method of retrieval-augmented text generation to generate a response to by an artificial intelligence (AI) agent. The method incudes constructing an indexing tree reflecting sematic relationship among textual contexts of one or more documents. The indexing tree includes at least a first level of nodes representing a plurality of propositions indicating facts about at least one entity from the documents, a second level of nodes representing a plurality of proposition aggregates that aggregate one or more first-level propositions mentioning a same entity, and a third level of nodes representing a plurality of summaries that summarize one or more second-level proposition aggregates. The method also includes receiving the user query; retrieving, from the indexing tree, one or more nodes across multiple levels based on similarity comparison with the user query based at least on a similarity between embeddings of the nodes and an embedding of the user query.

## Claims

### Claim 1

A method of retrieval-augmented text generation to generate a response to a user query by an artificial intelligence (AI) agent, the method comprising:
constructing, by one or more neural network based language models, an indexing tree reflecting sematic relationship among textual contexts of one or more documents, wherein the indexing tree comprises at least a first level of nodes representing a plurality of propositions indicating facts about at least one entity from the one or more documents, a second level of nodes representing a plurality of proposition aggregates that aggregate one or more first-level propositions mentioning a same entity, and a third level of nodes representing a plurality of summaries that summarize one or more second-level proposition aggregates;
receiving, via a communication interface, the user query;
retrieving, from the indexing tree, one or more nodes across multiple levels based on similarity comparison with the user query based at least on a similarity between embeddings of the one or more nodes and an embedding of the user query;
generating, by the neural network based language model, a response in response to the user query based on a context combining at least summaries and/or proposition aggregates corresponding to the retrieved one or more nodes; and
causing the response to be displayed at a visualized user interface of the AI agent.

### Claim 2 (depends on 1)

The method of claim 1, wherein the constructing the indexing tree comprises:
dividing or rewriting the one or more documents into a plurality of segments; and
extracting a plurality of entities and the plurality of propositions each relating to one or more entities from the plurality of segments.

### Claim 3 (depends on 2)

The method of claim 2, wherein the constructing of the indexing tree further comprises:
determining a first proposition and a second proposition on the first level relate to the same entity;
generating a first proposition aggregate based on the first proposition and the second proposition; and
adding a second-level parent node that is linked to two child nodes of the first proposition and the second proposition on the first level.

### Claim 4 (depends on 3)

The method of claim 3, wherein the constructing of the indexing tree further comprises:
generating a summary of the one or more second-level proposition aggregates that are clustered based on similarity; and
adding a third-level parent node of the summary linking to nodes representing the one or more second-level proposition aggregates.

### Claim 5 (depends on 3)

The method of claim 3, wherein the constructing of the indexing tree further comprises:
generating a summary of the one or more second-level proposition aggregates and one or more segments that are clustered based on similarity.

### Claim 6 (depends on 1)

The method of claim 1, wherein the retrieving the one or more nodes comprises:
retrieving, from the indexing tree, one or more summaries from the third level based on similarity comparison of the one or more summaries and the query embedding;
iteratively selecting, for each summary, one or more child nodes of the respective summary on the indexing tree based on similarity comparison of the respective child nodes and the query embedding;
traversing the index tree from a top level to lower levels until the second level of proposition aggregates; and
concatenating all retrieved nodes to form the context.

### Claim 7 (depends on 1)

The method of claim 1, wherein the retrieving the one or more nodes comprises:
retrieving, from a set of all summaries and proposition aggregates in the indexing tree, one or more summaries and/or proposition aggregates subject to a threshold number of tokens based on a similarity comparison between the set of all summaries and proposition aggregates, and an embedding of the user query, respectively,
wherein the set of all summaries and proposition aggregates in the indexing tree is obtained by flattening the indexing tree into a single layer.

### Claim 8

A system for retrieval-augmented text generation to generate a response to a user query by an artificial intelligence (AI) agent, the system comprising:
a memory that stores one or more neural network based language models and a plurality of processor executable instructions;
a communication interface that receives the user query; and
one or more hardware processors that read and execute the plurality of processor-executable instructions from the memory to perform operations comprising:
constructing, by one or more neural network based language models, an indexing tree reflecting sematic relationship among textual contexts of one or more documents, wherein the indexing tree comprises at least a first level of nodes representing a plurality of propositions indicating facts about at least one entity from the one or more documents, a second level of nodes representing a plurality of proposition aggregates that aggregate one or more first-level propositions mentioning a same entity, and a third level of nodes representing a plurality of summaries that summarize one or more second-level proposition aggregates;
receiving, via a communication interface, the user query;
retrieving, from the indexing tree, one or more nodes across multiple levels based on similarity comparison with the user query based at least on a similarity between embeddings of the one or more nodes and an embedding of the user query;
generating, by the neural network based language model, a response in response to the user query based on a context combining at least summaries and/or proposition aggregates corresponding to the retrieved one or more nodes; and
causing the response to be displayed at a visualized user interface of the AI agent.

### Claim 9 (depends on 8)

The system of claim 8, wherein the constructing the indexing tree comprises:
dividing or rewriting the one or more documents into a plurality of segments; and
extracting a plurality of entities and the plurality of propositions each relating to one or more entities from the plurality of segments.

### Claim 10 (depends on 9)

The system of claim 9, wherein the constructing of the indexing tree further comprises:
determining a first proposition and a second proposition on the first level relate to the same entity;
generating a first proposition aggregate based on the first proposition and the second proposition; and
adding a second-level parent node that is linked to two child nodes of the first proposition and the second proposition on the first level.

### Claim 11 (depends on 10)

The system of claim 10, wherein the constructing of the indexing tree further comprises:
generating a summary of the one or more second-level proposition aggregates that are clustered based on similarity; and
adding a third-level parent node of the summary linking to nodes representing the one or more second-level proposition aggregates.

### Claim 12 (depends on 10)

The system of claim 10, wherein the constructing of the indexing tree further comprises:
generating a summary of the one or more second-level proposition aggregates and one or more segments that are clustered based on similarity.

### Claim 13 (depends on 8)

The system of claim 8, wherein the retrieving the one or more nodes comprises:
retrieving, from the indexing tree, one or more summaries from the third level based on similarity comparison of the one or more summaries and the query embedding;
iteratively selecting, for each summary, one or more child nodes of the respective summary on the indexing tree based on similarity comparison of the respective child nodes and the query embedding;
traversing the index tree from a top level to lower levels until the second level of proposition aggregates; and
concatenating all retrieved nodes to form the context.

### Claim 14 (depends on 8)

The system of claim 8, wherein the retrieving the one or more nodes comprises:
retrieving, from a set of all summaries and proposition aggregates in the indexing tree, one or more summaries and/or proposition aggregates subject to a threshold number of tokens based on a similarity comparison between the set of all summaries and proposition aggregates, and an embedding of the user query, respectively,
wherein the set of all summaries and proposition aggregates in the indexing tree is obtained by flattening the indexing tree into a single layer.

### Claim 15

A non-transitory machine-readable medium comprising a plurality of machine-executable instructions which, when executed by one or more processors, are adapted to cause the one or more processors to perform operations comprising:
constructing, by one or more neural network based language models, an indexing tree reflecting sematic relationship among textual contexts of one or more documents, wherein the indexing tree comprises at least a first level of nodes representing a plurality of propositions indicating facts about at least one entity from the one or more documents, a second level of nodes representing a plurality of proposition aggregates that aggregate one or more first-level propositions mentioning a same entity, and a third level of nodes representing a plurality of summaries that summarize one or more second-level proposition aggregates;
receiving, via a communication interface, the user query;
retrieving, from the indexing tree, one or more nodes across multiple levels based on similarity comparison with the user query based at least on a similarity between embeddings of the one or more nodes and an embedding of the user query;
generating, by the neural network based language model, a response in response to the user query based on a context combining at least summaries and/or proposition aggregates corresponding to the retrieved one or more nodes; and
causing the response to be displayed at a visualized user interface of the AI agent.

### Claim 16 (depends on 15)

The method of claim 15, wherein the constructing the indexing tree comprises:
dividing or rewriting the one or more documents into a plurality of segments; and
extracting a plurality of entities and the plurality of propositions each relating to one or more entities from the plurality of segments.

### Claim 17 (depends on 16)

The method of claim 16, wherein the constructing of the indexing tree further comprises:
determining a first proposition and a second proposition on the first level relate to the same entity;
generating a first proposition aggregate based on the first proposition and the second proposition; and
adding a second-level parent node that is linked to two child nodes of the first proposition and the second proposition on the first level.

### Claim 18 (depends on 17)

The method of claim 17, wherein the constructing of the indexing tree further comprises:
generating a summary of the one or more second-level proposition aggregates that are clustered based on similarity; and
adding a third-level parent node of the summary linking to nodes representing the one or more second-level proposition aggregates.

### Claim 19 (depends on 17)

The method of claim 17, wherein the constructing of the indexing tree further comprises:
generating a summary of the one or more second-level proposition aggregates and one or more segments that are clustered based on similarity.

### Claim 20 (depends on 15)

The method of claim 15, wherein the retrieving the one or more nodes comprises:
retrieving, from the indexing tree, one or more summaries from the third level based on similarity comparison of the one or more summaries and the query embedding;
iteratively selecting, for each summary, one or more child nodes of the respective summary on the indexing tree based on similarity comparison of the respective child nodes and the query embedding;
traversing the index tree from a top level to lower levels until the second level of proposition aggregates; and
concatenating all retrieved nodes to form the context.
