GenerativeConjoint

An open-source, self-hosted conjoint survey platform with generative AI stimuli

GenerativeConjoint covers the full lifecycle of a conjoint study: survey design, D-optimal design generation, conventional tabular presnetations and AI-generated stimuli (text and images), participant-facing survey delivery, and export of analysis-ready data with live feedback. It is built for researchers who need full control over their data and infrastructure without commercial licensing costs and simplifies research data management through transparent exports of the data, the stimuli, and the stimuli generating prompts.

Example study and data can be found here: https://osf.io/6cqhx/overview

Features

Survey design

• Guided step-by-step wizard with attributes, levels, participant-facing descriptions, and LLM hints
• AI-assisted attribute and level suggestion from a survey title and description (research should be theory driven!)
• Standard and blocked designs
• Heuristic based sample-size estimation
• D-optimal design generation via coordinate exchange (20 random starts)
• Drag-and-drop reordering of attributes and levels
• Multi-user collaboration (owner / collaborator roles, ORCID-based invite)

Stimuli generation

• Tabular --- conventional attribute × alternative table, no AI required
• Textual --- LLM-generated scenario descriptions; one text per unique profile, shared across participants; level-specific LLM hints enrich generation without being shown to participants
• Visual --- AI-generated images (gpt-image-2, gpt-image-1, gpt-image-1-mini, or DALL-E 3); configurable quality (low / medium / high); one image per unique profile
• Per-stimulus regeneration from the Inspect Design page
• Full audit trail: prompts stored alongside outputs

Participant experience

• Entry via unique URL with external participant key (?key=KEY); auto-generates an anonymous session key if no key is supplied
• Optional introduction page with consent checkbox (AI generatable)
• Randomisable attribute order (per-participant or per-task)
• Response-time tracking
• Optional auto-advance and configurable redirect URL on completion

Results and export

• Results dashboard: completion stats and response time, part-worth utility estimates
• Shareable results link (token-protected, no login required)
• ZIP export bundle:
  • responses.csv --- long-format responses (one row per participant × task × alternative)
  • design_matrix.csv --- the fixed conjoint design
  • design.yaml --- full study design including attributes, levels, descriptions, and LLM hints
  • description.md --- human-readable data dictionary and methodology notes
  • generated_texts.csv / generated_images.zip --- all AI-generated stimuli with prompts
  • r_starter.R --- ready-to-run R script (clogit / mlogit examples)

---

Tech stack

Layer	Technology
Backend	Python 3.9+, Flask 3, SQLAlchemy
Database	SQLite (default) or PostgreSQL
AI	OpenAI API (chat completions + image generation)
Frontend	Bootstrap 5.3, SortableJS, vanilla JS
Auth	ORCID OAuth 2.0

---

Getting started

1. Clone and install

git clone https://github.com/braunerphilipp/GenerativeConjoint.git
cd GenerativeConjoint
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Configure

Copy the example environment file and fill in your credentials:

cp .env.example .env

Variable	Description
SECRET_KEY	Long random string for session signing
DATABASE_URL	SQLAlchemy URI (default: sqlite:///conjoint.db)
OPENAI_API_KEY	OpenAI API key for text and image generation
ORCID_CLIENT_ID	ORCID OAuth app client ID
ORCID_CLIENT_SECRET	ORCID OAuth app secret
ORCID_REDIRECT_URI	OAuth callback URL (e.g. http://localhost:5000/auth/callback)
ORCID_BASE_URL	https://sandbox.orcid.org (dev) or https://orcid.org (prod)
ORCID_API_URL	https://pub.sandbox.orcid.org (dev) or https://pub.orcid.org (prod)

Register an OAuth application at orcid.org/developer-tools to obtain the ORCID credentials.

3. Run

python app.py

The app starts at http://localhost:5000. The database is created automatically on first run; schema migrations are applied in-place on startup.

---

Deployment (Apache + mod_wsgi)

Create wsgi.py in the project root:

import sys, os
sys.path.insert(0, '/path/to/GenerativeConjoint')
from app import app as application

Add to your Apache VirtualHost:

WSGIDaemonProcess conjoint python-home=/path/to/venv \
    python-path=/path/to/GenerativeConjoint threads=4
WSGIProcessGroup conjoint
WSGIScriptAlias / /path/to/GenerativeConjoint/wsgi.py

Alias /static /path/to/GenerativeConjoint/static

---

Citation

If you use GenerativeConjoint in your research, please cite:

Brauner, P. (2025). GenerativeConjoint: An Open-Source Conjoint Survey Platform with
Generative AI Stimuli.

---

Acknowledgements

Developed by Philipp Brauner, Communication Science, RWTH Aachen University.