CAISI Research Catalog

A curated catalog of AI safety research implementations and tools developed under the Canadian AI Safety Institute (CAISI) Research Program at CIFAR.

Web Interface

The catalog is an interactive static website featuring:

• Fast Search — instant full-text search powered by Pagefind
• Filtering — browse by type (applied-research, bootcamp, tool)
• Responsive Design — works on all devices
• Static Deployment — zero backend, deployable to GCS or GitHub Pages

Local Development

# 1. Install Python dependencies and sync repositories data
pip install pyyaml
python scripts/sync_repositories_to_json.py

# 2. Build and serve the catalog
cd catalog
npm install
npm run build   # builds Next.js + generates search index
npm run serve   # serves the static output on http://localhost:3001

For hot-reload during frontend development (search won't work without a build):

cd catalog
npm run dev

Adding a Repository

Create a new YAML file in repositories/ named <repo-name>.yaml:

name: my-safety-tool              # Required: short identifier
repo_id: VectorInstitute/my-repo  # Required: GitHub org/repo
description: "Brief description"  # Required
implementations:                  # Required: list of algorithms/techniques
  - name: Safety Evaluation Suite
    url: null                     # Optional URL to specific implementation
type: tool                        # Required: tool | bootcamp | applied-research
year: 2025                        # Required: publication/release year
github_url: https://...           # Optional: overrides repo_id URL
paper_url: https://...            # Optional: associated paper link
bibtex: citation-key              # Optional: key in catalog/public/data/papers.bib
platform_url: https://...         # Optional: deployment platform URL
public_datasets:                  # Optional: datasets used
  - name: Dataset Name
    url: https://...

After adding a YAML file, regenerate the JSON:

python scripts/sync_repositories_to_json.py

Required Fields

Field	Type	Description
name	string	Repository identifier
repo_id	string	GitHub org/repo
description	string	Brief description
implementations	list	ML/AI techniques implemented
type	string	tool, bootcamp, or applied-research
year	integer	Publication/release year

GitHub Actions

Workflow	Trigger	Purpose
code_checks.yml	Push/PR	Runs ruff, typos, Next.js lint, pip-audit
collect-metrics.yml	Weekly (Mon 12:00 UTC)	Collects GitHub stars/forks/traffic
deploy-catalog-gcp.yml	Push to main / metrics update	Builds and deploys to GCS

Required Secrets

For deployment, configure these GitHub Actions secrets:

Secret	Description
GCP_WORKLOAD_IDENTITY_PROVIDER	GCP Workload Identity Federation provider
GCP_SERVICE_ACCOUNT	GCP service account email
GCS_BUCKET	Target GCS bucket name
METRICS_GITHUB_TOKEN	PAT with repo scope for traffic data