CoolFace
Apppublic

cpulido/NDE-NARRATIVES-ANALYSIS

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face
1likes
App README

NDE Narratives Analysis (Public Demo)

Welcome. This Space is a public, lightweight demo of the NDE narratives project.

It is designed so anyone can quickly test how near-death narrative text can be transformed into structured outputs.

Project Context

This demo is based on a broader research line studying representational mismatch in near-death experiences (NDEs):

  • —people can report an NDE in free narrative form (rich, contextual, temporal),
  • —and also answer structured questionnaires (standardized labels),
  • —but these two representations do not always align perfectly.

The core research objective is to quantify that partial alignment and understand where mismatch is systematic.

In the full study workflow, narratives are analyzed in three sections:

  1. 1.Context (before/around the event)
  2. 2.Experience (the core NDE)
  3. 3.Aftereffects (longer-term changes)

This public Space mirrors that logic at small scale for interactive exploration.

Video Summary

Short overview of article-aligned findings and the narrative-vs-questionnaire framing:

<p align="center"> <video src="https://raw.githubusercontent.com/cristian-pulido/NDENARRATIVESANALYSIS/main/StoriesvsSurveys.mov" controls width="760"></video> </p>

Fallback direct link: `Stories_vs_Surveys.mov`

What This Space Includes

  • —Three-section analysis mode only (context, experience, aftereffects all required)
  • —Stage-based workflow view (input -> segmentation -> module analysis -> interpretation)
  • —Module tabs for tone estimation, structured features, and alignment layer
  • —Optional valence-vs-tone alignment message with evidence tracing
  • —Default local model option (Qwen 3.5 0.8B) loaded in-instance after first download
  • —Additional routed models enabled in this Space: Qwen/Qwen2.5-7B-Instruct, deepseek-ai/DeepSeek-R1-Distill-Llama-8B

Prompt Alignment with Local Version

This Space uses the same section-prompt logic as the main local workflow (context, experience, aftereffects), adapted for both local and routed inference.

Runtime note:

  • —Default local model downloads once and then runs from instance cache (/tmp/nde_hf_cache).
  • —HF_TOKEN is recommended for both local model download reliability and routed inference access.
  • —Free usage depends on your Hugging Face account credits/quota and current provider availability.

Optional override:

  • —Set LOCAL_QWEN_MODEL_ID in Space Variables if you want to change the local default model id.

Why This Public Version Is Simplified

This Space runs on free/shared Hugging Face infrastructure and uses a simplified configuration around a single lightweight model (Qwen/Qwen3.5-0.8B) for easy public access.

To keep this demo easy to run and publicly accessible, it replaces heavier local tooling with a reduced setup.

That means this Space is for:

  • —quick testing,
  • —educational exploration,
  • —and transparent prototyping.

It is not the final research-grade execution environment.

Full Local Research Version (Recommended)

For the more solid, reproducible workflow (including local Ollama models such as Gemma/Qwen, full preprocessing, experiment management, and evaluation), use:

  • —https://github.com/cristian-pulido/NDENARRATIVESANALYSIS

Important Disclaimer

  • —Research and educational use only; not a medical or psychological diagnostic tool.
  • —Do not submit personally identifying or highly sensitive information.
  • —Model outputs may be inaccurate, incomplete, or biased and require human review.