How tabiya-tech/compass works
Project type and purpose
• This is a full-stack AI chatbot (“Compass”) for job seekers, comprising:
– A Python/FastAPI-based backend that orchestrates LLM calls, vector search, state management and multiple domain services (feedback, invitations, metrics, sensitive-data filtering, etc.).
– Two frontends:
• “frontend” – a legacy Next.js proof-of-concept chat UI
• “frontend-new” – the production-grade React/Next.js TypeScript app with Firebase-based auth and a rich component library
– An IaC layer (Pulumi scripts) that deploys all services to AWS (and some GCP identity resources).
File structure
• Top level:
– backend/ – Python service
– frontend-new/ – Next.js α React UI
– frontend/ – POC chat UI
– iac/ – Pulumi infrastructure code
• backend/
– app/ – core; server.py (FastAPI entrypoint), modules for agent orchestration, conversation memory, vector search, user/feedback/invitations/metrics routes, persistence stores
– common_libs/ – shared utilities (LLM callers, logging, retry, test helpers)
– evaluation_tests/ – end-to-end and unit tests for each agent and pipeline
– scripts/ – maintenance and data-prep scripts (embeddings, migrations, export/import)
• frontend-new/
– src/app/ – global wrappers (storage, routing, viewport)
– src/auth/ – Firebase config, service factories, auth UI pages
– src/chat/ – chat components + ChatService wrapper for API calls
– src/… – experiences drawer, reports, feedback forms, metrics, sensitive-data forms, user preferences, theming, utilities, global error handling
• iac/
– subfolders by stack (auth, aws-ns, backend, frontend, dns, environment, realm, common)
– each contains Pulumi.yaml + Python deploy scripts
Parse the README for additional insights
• Compass helps job-seekers extract skills and experiences via AI agents.
• Two frontend flavors exist until the new UI is complete.
• CI/CD is defined in .github/workflows for backend, frontend, config and deploy.
• run-before-merge.sh orchestrates lint/build/test across all subprojects.
Detail level
– Don’t omit any persistence store (MongoDB, JSON store, vector store)
– Include the IAC pipelines feeding into the deployment of front/back