PulseCX: Breaking the Closed-World Assumption in Real-Time CX

Rajat Agarwal; Shubham Sharma; Suvidha Tripathi

arxiv: 2606.21124 · v1 · pith:PGURT6MHnew · submitted 2026-06-19 · 💻 cs.AI · cs.IR

PulseCX: Breaking the Closed-World Assumption in Real-Time CX

Rajat Agarwal , Suvidha Tripathi , Shubham Sharma This is my paper

Pith reviewed 2026-06-26 14:41 UTC · model grok-4.3

classification 💻 cs.AI cs.IR

keywords conversational AIcustomer experienceclosed-world assumptiontemporal knowledge graphasynchronous agentsintent resolutionreal-time systemsdecay dynamics

0 comments

The pith

PulseCX decouples knowledge gathering from use in CX agents via an asynchronous decay-aware graph to handle external shifts without added latency.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Conversational agents for customer experience typically ignore fast external changes such as viral trends or outages because they operate under a closed-world assumption. Ad-hoc searches to fix this add latency and risk feeding irrelevant or harmful context into responses. PulseCX separates acquisition of new signals from their consumption in answers. An asynchronous agent turns incoming signals into a Decay-Aware Temporal Knowledge Graph whose nodes and edges follow reinforcement-decay rules that control how long each piece of information remains active. Pairing this memory with hierarchical intent gating keeps added overhead below 10 milliseconds while raising intent resolution rates and satisfaction scores in changing conditions.

Core claim

PulseCX is a framework that decouples knowledge acquisition from consumption. It employs an asynchronous agent to linearize external signals into a Decay-Aware Temporal Knowledge Graph (DA-TKG) governed by reinforcement-decay dynamics to actively manage information lifecycles. By coupling this self-evolving memory with hierarchical intent gating, PulseCX removes synchronous search bottlenecks with less than 10 ms overhead and drives gains in Intent Resolution Rate and customer satisfaction metrics in dynamic environments.

What carries the argument

The Decay-Aware Temporal Knowledge Graph (DA-TKG) governed by reinforcement-decay dynamics, which acts as a self-evolving memory that linearizes external signals asynchronously and controls information lifespan.

If this is right

Synchronous web searches are no longer required, removing their latency and poisoning risks.
Intent Resolution Rate and satisfaction scores improve in environments with rapid external change.
Information lifecycles are managed automatically so outdated signals do not persist.
The same memory structure supports multiple intents without rebuilding context on each turn.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same decoupling pattern could apply to other real-time agents that must track fast-moving external facts, such as financial news bots or public-health chat systems.
Automatic decay rules might reduce the engineering effort needed to keep any long-running agent memory current over weeks or months.
If the linearization step proves accurate, the approach could be combined with existing retrieval systems rather than replacing them outright.

Load-bearing premise

An asynchronous agent can reliably linearize external signals into a DA-TKG governed by reinforcement-decay dynamics while avoiding context poisoning and keeping overhead under 10 ms when paired with hierarchical intent gating.

What would settle it

A live test in which external signals during a viral trend produce context poisoning in the DA-TKG or push response latency above 10 ms would falsify the performance claims.

Figures

Figures reproduced from arXiv: 2606.21124 by Rajat Agarwal, Shubham Sharma, Suvidha Tripathi.

**Figure 1.** Figure 1: PulseCX Architecture: Offline Stage: Asynchronous discovery and world modeling via DA-TKG memory evolution. Online Stage: Real-time intent-aware grounding and feedback-driven refinement 2 RELATED WORK Agent Memory. Prior work on memory for LLM-based agents emphasizes short-term, episodic, or long-term recall for coherence and planning, typically assuming a static external world Wang et al. (2024); Hu et al… view at source ↗

read the original abstract

Conversational AI agents in Customer Experience (CX) typically suffer from a Closed-World Constraint, ignoring high-velocity external shifts like viral trends or outages. Ad-hoc web search attempts to bridge this gap but often introduce prohibitive latency and context poisoning. We introduce PulseCX, a framework that decouples knowledge acquisition from consumption. Adopting a structure-first paradigm, PulseCX employs an asynchronous agent to linearize signals into a Decay-Aware Temporal Knowledge Graph (DA-TKG) governed by reinforcement--decay dynamics to actively manage information lifecycles. By coupling this self-evolving memory with hierarchical intent gating, PulseCX removes synchronous search bottlenecks (<10ms overhead) and drives significant gains in Intent Resolution (IRR) and Customer Satisfaction (s-CSAT) in dynamic environments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

PulseCX sketches a plausible architecture for async knowledge handling in CX agents but states its key performance numbers with zero supporting data or experiments.

read the letter

The main takeaway is that this paper describes a framework called PulseCX that tries to let conversational CX agents pull in external signals without the usual latency or poisoning problems, using an async agent to build a decay-aware temporal knowledge graph under reinforcement-decay rules plus hierarchical intent gating. The core move—decoupling acquisition from consumption—is a clean way to think about real-time memory in agents.

What stands out as new is the specific bundle of decay-aware TKG, reinforcement-decay dynamics, and the gating layer; the abstract frames this as a structure-first approach to information lifecycles. The motivation around closed-world limits and ad-hoc search failures is stated plainly.

The soft spot is straightforward and large: the abstract asserts <10 ms overhead plus significant lifts in IRR and s-CSAT, yet supplies no measurements, baselines, ablations, scaling data, or even a description of how the system was run. There is no evidence that the linearization step stays fast or that context poisoning is actually controlled. Without those pieces the quantitative claims sit as assertions rather than results.

The paper is aimed at applied researchers building real-time conversational systems who might want the high-level pattern. A reader looking for concrete methods or reproducible gains will find little to use. I would not send this to peer review until the authors add at least basic experiments and measurements; the idea is coherent enough on paper but the current version does not yet demonstrate it works.

Referee Report

2 major / 1 minor

Summary. The paper introduces PulseCX, a framework for conversational AI in customer experience (CX) that decouples knowledge acquisition from consumption via an asynchronous agent linearizing external signals into a Decay-Aware Temporal Knowledge Graph (DA-TKG) governed by reinforcement-decay dynamics. Coupled with hierarchical intent gating, the approach is claimed to eliminate synchronous search bottlenecks with <10ms overhead while delivering significant gains in Intent Resolution Rate (IRR) and customer satisfaction (s-CSAT) in dynamic environments.

Significance. If validated, the structure-first paradigm and self-evolving memory could meaningfully advance real-time CX agents by addressing the closed-world constraint without prohibitive latency or context poisoning. The decoupling of acquisition and consumption represents a potentially useful architectural shift for handling high-velocity external signals.

major comments (2)

[Abstract] Abstract: the central claims of significant IRR/s-CSAT gains and <10ms overhead are asserted without any experiments, baselines, latency measurements, ablation studies, or error analysis. No data or formal argument is supplied to support that the DA-TKG linearization plus gating actually achieves these outcomes or avoids context poisoning.
[Abstract] Abstract: the DA-TKG, reinforcement-decay dynamics, and hierarchical intent gating are introduced as invented entities without definitions, equations, pseudocode, or complexity analysis, rendering the performance assertions impossible to evaluate or reproduce.

minor comments (1)

The acronym s-CSAT is used without expansion or definition.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the opportunity to respond to the referee's report. We appreciate the detailed feedback on the abstract and will revise the manuscript to address the concerns.

read point-by-point responses

Referee: [Abstract] Abstract: the central claims of significant IRR/s-CSAT gains and <10ms overhead are asserted without any experiments, baselines, latency measurements, ablation studies, or error analysis. No data or formal argument is supplied to support that the DA-TKG linearization plus gating actually achieves these outcomes or avoids context poisoning.

Authors: We agree that the abstract asserts quantitative performance claims without supporting evidence in the provided text. We will revise the manuscript to add a dedicated evaluation section containing experiments, baselines, latency measurements confirming the overhead, ablation studies, and error analysis to substantiate the IRR, s-CSAT, and context-poisoning claims. revision: yes
Referee: [Abstract] Abstract: the DA-TKG, reinforcement-decay dynamics, and hierarchical intent gating are introduced as invented entities without definitions, equations, pseudocode, or complexity analysis, rendering the performance assertions impossible to evaluate or reproduce.

Authors: We agree that the abstract introduces the key components without definitions or technical details. We will revise the manuscript to include formal definitions, equations for the reinforcement-decay dynamics, pseudocode for the hierarchical intent gating, and complexity analysis. revision: yes

Circularity Check

0 steps flagged

No significant circularity: framework description contains no derivation chain or self-referential reductions

full rationale

The provided abstract and description introduce PulseCX as a framework employing an asynchronous agent to build a DA-TKG with reinforcement-decay dynamics and hierarchical intent gating, asserting <10ms overhead and gains in IRR/s-CSAT. No equations, fitted parameters, uniqueness theorems, or self-citations appear in the text. Without any load-bearing derivation steps, mathematical claims, or ansatzes that could reduce to inputs by construction, the paper's presentation is self-contained as a high-level system proposal rather than a derived result. The performance assertions remain unsupported by evidence but do not exhibit circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Review performed on abstract only; full text unavailable. Ledger therefore contains only the minimal assumptions extractable from the abstract.

axioms (1)

domain assumption External high-velocity signals can be linearized into a temporal knowledge graph without loss of utility for downstream intent resolution.
Invoked by the structure-first paradigm and DA-TKG construction.

invented entities (1)

Decay-Aware Temporal Knowledge Graph (DA-TKG) no independent evidence
purpose: Self-evolving memory that manages information lifecycles via reinforcement-decay dynamics.
New construct introduced to decouple acquisition from consumption.

pith-pipeline@v0.9.1-grok · 5657 in / 1303 out tokens · 16739 ms · 2026-06-26T14:41:43.697187+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

19 extracted references · 5 canonical work pages · 1 internal anchor

[1]

Scaling Learning Algorithms Towards

Bengio, Yoshua and LeCun, Yann , booktitle =. Scaling Learning Algorithms Towards
[2]

and Osindero, Simon and Teh, Yee Whye , journal =

Hinton, Geoffrey E. and Osindero, Simon and Teh, Yee Whye , journal =. A Fast Learning Algorithm for Deep Belief Nets , volume =
[3]

2016 , publisher=

Deep learning , author=. 2016 , publisher=

2016
[4]

arXiv preprint arXiv:2512.05470 , year=

Everything is context: Agentic file system abstraction for context engineering , author=. arXiv preprint arXiv:2512.05470 , year=

work page arXiv
[5]

Proceedings of the ACM on Management of Data , volume=

CtxPipe: Context-aware Data Preparation Pipeline Construction for Machine Learning , author=. Proceedings of the ACM on Management of Data , volume=. 2024 , publisher=

2024
[6]

Advances in Neural Information Processing Systems , volume=

Agentpoison: Red-teaming llm agents via poisoning memory or knowledge bases , author=. Advances in Neural Information Processing Systems , volume=
[7]

arXiv preprint arXiv:2512.17041 , year=

Security Risks of Agentic Vehicles: A Systematic Analysis of Cognitive and Cross-Layer Threats , author=. arXiv preprint arXiv:2512.17041 , year=

work page arXiv
[8]

Memory in the Age of AI Agents

Memory in the Age of AI Agents , author=. arXiv preprint arXiv:2512.13564 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[9]

International Conference on Computational Science and Computational Intelligence , pages=

Revolutionizing Multi-agent Systems: The Role of Agentic RAG in Dynamic Data Ingestion and Real-Time Reasoning , author=. International Conference on Computational Science and Computational Intelligence , pages=. 2024 , organization=

2024
[10]

Frontiers of Computer Science , volume=

A survey on large language model based autonomous agents , author=. Frontiers of Computer Science , volume=. 2024 , publisher=

2024
[11]

NeurIPS , year=

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks , author=. NeurIPS , year=
[12]

ICLR , year=

ReAct: Synergizing Reasoning and Acting in Language Models , author=. ICLR , year=
[13]

Journal of Business and Future Economy , volume=

Enhancing Customer Experience and Business Intelligence: The Role of AI-Driven Smart CRM in Modern Enterprises , author=. Journal of Business and Future Economy , volume=
[14]

Advances in digital marketing in the era of artificial intelligence , pages=

Enhancing customer experience through AI-enabled content personalization in e-commerce marketing , author=. Advances in digital marketing in the era of artificial intelligence , pages=. 2024 , publisher=

2024
[15]

, author=

Context Aware Intelligence: A Framework for Immersive Customer Experience. , author=. International Journal of Advanced Research in Computer Science , volume=
[16]

Sustainability , volume=

Customer experience design for smart product-service systems based on the iterations of experience--evaluate--engage using customer experience data , author=. Sustainability , volume=. 2022 , publisher=

2022
[17]

arXiv preprint arXiv:2507.13396 , year=

DyG-RAG: Dynamic Graph Retrieval-Augmented Generation with Event-Centric Reasoning , author=. arXiv preprint arXiv:2507.13396 , year=

work page arXiv
[18]

arXiv preprint arXiv:2402.16568 , year=

Two-stage generative question answering on temporal knowledge graph using large language models , author=. arXiv preprint arXiv:2402.16568 , year=

work page arXiv
[19]

Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

Temporal knowledge question answering via abstract reasoning induction , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

[1] [1]

Scaling Learning Algorithms Towards

Bengio, Yoshua and LeCun, Yann , booktitle =. Scaling Learning Algorithms Towards

[2] [2]

and Osindero, Simon and Teh, Yee Whye , journal =

Hinton, Geoffrey E. and Osindero, Simon and Teh, Yee Whye , journal =. A Fast Learning Algorithm for Deep Belief Nets , volume =

[3] [3]

2016 , publisher=

Deep learning , author=. 2016 , publisher=

2016

[4] [4]

arXiv preprint arXiv:2512.05470 , year=

Everything is context: Agentic file system abstraction for context engineering , author=. arXiv preprint arXiv:2512.05470 , year=

work page arXiv

[5] [5]

Proceedings of the ACM on Management of Data , volume=

CtxPipe: Context-aware Data Preparation Pipeline Construction for Machine Learning , author=. Proceedings of the ACM on Management of Data , volume=. 2024 , publisher=

2024

[6] [6]

Advances in Neural Information Processing Systems , volume=

Agentpoison: Red-teaming llm agents via poisoning memory or knowledge bases , author=. Advances in Neural Information Processing Systems , volume=

[7] [7]

arXiv preprint arXiv:2512.17041 , year=

Security Risks of Agentic Vehicles: A Systematic Analysis of Cognitive and Cross-Layer Threats , author=. arXiv preprint arXiv:2512.17041 , year=

work page arXiv

[8] [8]

Memory in the Age of AI Agents

Memory in the Age of AI Agents , author=. arXiv preprint arXiv:2512.13564 , year=

work page internal anchor Pith review Pith/arXiv arXiv

[9] [9]

International Conference on Computational Science and Computational Intelligence , pages=

Revolutionizing Multi-agent Systems: The Role of Agentic RAG in Dynamic Data Ingestion and Real-Time Reasoning , author=. International Conference on Computational Science and Computational Intelligence , pages=. 2024 , organization=

2024

[10] [10]

Frontiers of Computer Science , volume=

A survey on large language model based autonomous agents , author=. Frontiers of Computer Science , volume=. 2024 , publisher=

2024

[11] [11]

NeurIPS , year=

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks , author=. NeurIPS , year=

[12] [12]

ICLR , year=

ReAct: Synergizing Reasoning and Acting in Language Models , author=. ICLR , year=

[13] [13]

Journal of Business and Future Economy , volume=

Enhancing Customer Experience and Business Intelligence: The Role of AI-Driven Smart CRM in Modern Enterprises , author=. Journal of Business and Future Economy , volume=

[14] [14]

Advances in digital marketing in the era of artificial intelligence , pages=

Enhancing customer experience through AI-enabled content personalization in e-commerce marketing , author=. Advances in digital marketing in the era of artificial intelligence , pages=. 2024 , publisher=

2024

[15] [15]

, author=

Context Aware Intelligence: A Framework for Immersive Customer Experience. , author=. International Journal of Advanced Research in Computer Science , volume=

[16] [16]

Sustainability , volume=

Customer experience design for smart product-service systems based on the iterations of experience--evaluate--engage using customer experience data , author=. Sustainability , volume=. 2022 , publisher=

2022

[17] [17]

arXiv preprint arXiv:2507.13396 , year=

DyG-RAG: Dynamic Graph Retrieval-Augmented Generation with Event-Centric Reasoning , author=. arXiv preprint arXiv:2507.13396 , year=

work page arXiv

[18] [18]

arXiv preprint arXiv:2402.16568 , year=

Two-stage generative question answering on temporal knowledge graph using large language models , author=. arXiv preprint arXiv:2402.16568 , year=

work page arXiv

[19] [19]

Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

Temporal knowledge question answering via abstract reasoning induction , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=