Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

ECHO: A Locally-Deployable Agentic Health Assistant with Temporal Memory, Safety Guardrails, and Speech Assessment

Source

arXiv

arxiv.org/abs/2608.06110v1

Read the source

Source

Plain-Language Brief

The paper introduces ECHO, a locally-deployable health assistant designed for long-term chronic care management. It consists of three integrated modules: an agentic chatbot, a safety layer, and a speech assessment module. The chatbot uses persistent memory and 17 clinical tools to converse with users, achieving high accuracy with GPT-5 Mini in a benchmark. The safety layer quickly identifies crisis signals or attempts to bypass the system, while a graph neural network classifies ambiguous cases. The speech assessment module evaluates user's emotion, depression, and pain levels using Whisper and BERT. ECHO runs on consumer hardware and complies with GDPR and KVKK data privacy laws.

What the Document Says

  • ECHO is a locally-deployable health assistant for long-term chronic care management.
  • It consists of three modules: an agentic chatbot, a safety layer, and a speech assessment module.
  • The chatbot uses persistent memory and 17 clinical tools.
  • It achieves high accuracy with GPT-5 Mini in a benchmark.
  • The safety layer identifies crisis signals or attempts to bypass the system quickly.
  • A graph neural network classifies ambiguous cases.
  • The speech assessment module evaluates user's emotion, depression, and pain levels using Whisper and BERT.
  • ECHO runs on consumer hardware and complies with GDPR and KVKK.

Key Concepts Mentioned

  • Health Assistant
  • Chronic Care Management
  • Locally-Deployable
  • Agentic Chatbot
  • ReAct loop
  • LangGraph
  • Clinical Tools
  • Temporal Knowledge Graph
  • Safety Layer
  • Rule-Based Layer
  • Signed Graph Neural Network (GNN)
  • APPNP-style propagation
  • Boundary Cases
  • Clinical Intent
  • Accuracy
  • Unsafe Recall
  • Annotated Turkish Health Dataset
  • Zero-shot LLM baselines
  • Llama 3.3 70B
  • Multimodal Speech Assessment
  • Whisper Acoustic Encoding
  • BERT Text Encoding
  • Cross-Attention Fusion
  • Emotion
  • Depression
  • Pain
  • Consumer Hardware
  • GDPR
  • KVKK

Claims Made by the Document

  • ECHO achieves a 94.9% tool-execution pass rate across a 59-scenario benchmark with GPT-5 Mini.
  • The two-stage hybrid safety layer intercepts all incoming queries and achieves 88.8% accuracy and 90.6% unsafe recall on a 2,537-query annotated Turkish health dataset.
  • The multimodal speech assessment module reaches a mean macro F1 of 0.652.

Evidence Used

  • Results from a 59-scenario benchmark with GPT-5 Mini.
  • Results from a 2,537-query annotated Turkish health dataset.
  • Results from the multimodal speech assessment module.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

This paper presents ECHO (Enhanced Care & Health Observer), a locally-deployable conversational health assistant for long-term chronic care management. ECHO integrates three complementary software modules developed under shared supervision as a unified system. The core module is an agentic chatbot built on a ReAct loop orchestrated via LangGraph, equipped with 17 clinical tools and a temporal knowledge graph for persistent cross-session memory; it achieves a 94.9% tool-execution pass rate across a 59-scenario benchmark with GPT-5 Mini. A two-stage hybrid safety layer intercepts all incoming queries: a rule-based layer handles explicit crisis signals and jailbreak attempts in under 1ms, while a signed graph neural network (GNN) with APPNP-style propagation classifies boundary cases by clinical intent, achieving 88.8% accuracy and 90.6% unsafe recall on a 2,537-query annotated Turkish health dataset while outperforming zero-shot LLM baselines including Llama 3.3 70B. A multimodal speech assessment module combining Whisper acoustic encoding and BERT text encoding with cross-attention fusion estimates emotion, depression, and pain, reaching a mean macro F1 of 0.652. The full system is implemented as a web application that can run entirely on consumer hardware, with no patient data transmitted to external services, supporting compliance with GDPR and KVKK.

Links to this note 1