Verita is an AI-focused community platform where developers, researchers, and enthusiasts share and discover practical AI knowledge through intelligent curation, automated summarization, and personalized recommendations. It is built as four backend microservices (three Spring Boot, one FastAPI) behind a React frontend.

• Reviewing the project? Start with the Review Guide.
• Developing locally? See Local Development.
• Everything else: the documentation index below.

Architecture at a Glance

Full rationale in System Overview & Architecture and the ADRs.

Quick Start with Docker Compose (Local)

The fastest way to run the full platform.

First-time setup. Create genai-service/.env from the template:

cp genai-service/.env.example genai-service/.env

The placeholder values are enough to boot. The recommended setup for testing GenAI is Logos with openai/gpt-oss-120b, which is already the template default — add a real LOGOS_API_KEY in genai-service/.env to enable AI summaries and daily digest generation, or point the service at a local Ollama instead. A GNEWS_API_KEY is recommended for reliable digest source coverage, but is not required to run the service. See GenAI Environment Setup for the other providers and source API keys.

Run the docker compose from the repository root:

docker compose up --build

This builds the application services plus one PostgreSQL database per service, MinIO object storage, and a local mail sink:

Common lifecycle commands:

docker compose down                       # stop all services (data is kept)
docker compose down -v                    # stop and delete all data (DBs + object storage)
docker compose up --build user-service    # start a single service only

Database and MinIO connection details and credentials live in Local Development.

Seed Demo Data

A fresh stack starts empty. Once the services are healthy, seed the local databases with demo users, posts, topics, comments, bookmarks, votes, follows, and notifications:

npm install
npm run seed:local

The seed is idempotent and non-destructive; options and the full breakdown are in Local Development.

Then open http://localhost:3000, sign in as a demo account, and follow the Review Guide for a guided tour.

A Prometheus + Grafana stack can be layered on top — see the Review Guide to bring it up, and infra/monitoring/README.md for what is collected and the Azure/Kubernetes variants.

Cloud Deployments

Live environments:

Note: don't forget to type thisisunsafe when the browser warns about the self-signed TLS certificate on the dev environment.

See Infrastructure Design and the Helm Chart for how these are provisioned and deployed.

Demo Accounts

Authentication is always against the real backend. Seed the database (npm run seed:local) to create the demo users, then log in at http://localhost:3000. All seed users share the password Password123!; profiles, posts, bookmarks, and likes are populated by the seed.

Documentation

Project-wide documentation lives under docs/ — full index in docs/README.md. Component and tooling docs live next to the code they describe. Key entry points:

See It Running

Recorded against a real, seeded stack — every screen below is the running application, not a mockup. Clips are sped up to keep them short.

The core loop

Sign in against the real user-service, and the guest banner gives way to a personally ranked For you feed. Read a post, then like, save, share, and comment — every count moves against the live backend.

The daily digest

One digest per day, the reader's own badged Personalized. It is assembled from Hugging Face, GNews, and GitHub, then written by the LLM — the clip follows a citation out to the real release page behind it.

Authoring

Write in Markdown and watch it render live, KaTeX math included. Add sources and tags, publish, and the post goes straight to the AI for summarizing.

Admin: operating the AI

The Operations panel reads active: back from GenAI and can switch the provider and model at runtime. Re-summarize a real post, watch the job land COMPLETED against a live LLM, then read the summary it just wrote — stamped with the model behind it.