A multi-tenant AI data platform built from the ground up
This repository is my full-stack demonstration project: a production‑grade system that combines LLMs, hybrid RAG retrieval, text‑to-SQL analytics, multi‑tenant security, and human‑in‑the‑loop workflows.
It showcases my skills across backend development, AI orchestration, frontend engineering, infra, and observability.
- Full-stack architecture: React 18 dashboard + FastAPI microservices + LangGraph agents.
- AI-centric design: Hybrid RAG search, provider-agnostic LLM service (OpenAI, Anthropic, Google, Ollama) with tiered budgets.
- Multi-tenancy at the DB layer: PostgreSQL Row-Level Security, Supabase Edge Functions, and per-tenant context injection.
- Text-to-SQL analytics: Natural language queries converted to safe SQL with allowlists, PII masking, and RLS enforcement.
- Human-in-the-Loop quality control: Custom review queue and elicitation service with Streamlit UI.
- Tier-based access control: Six granular tiers (FREE→ADMIN) governing tool availability and rate limits.
- Observability-first: OpenTelemetry, Langfuse traces, Grafana Loki/Tempo/Mimir/Faro.
- Extensible libraries: 20+ shared Python packages used by all services (zero duplication).
- Deployment-ready: Docker Compose for dev, Google Cloud Run for production with CI-ready commands.
This project synthesizes dozens of technologies and demonstrates solid engineering practices, including clean architecture, automated testing, linting, and detailed documentation.
- Python 3.11+, FastAPI, Pydantic, SQLModel
- Building and orchestrating LLM-based agents (LangGraph, LangChain)
- Designing multi-tenant systems with PostgreSQL RLS
- Developing hybrid retrieval pipelines (pgvector + Cohere + MMR)
- Implementing secure text‑to‑SQL with allowlists, masking, and query sanitization
- Creating interactive, hot‑reload React applications (Vite, Chakra UI)
- Working with Supabase (Edge Functions, Auth, pgvector)
- Managing Redis for caching, checkpoints, and prioritized queues
- Integrating observability (OpenTelemetry, Langfuse, Grafana Cloud)
- Containerization and deployment (Docker, Cloud Run, Artifact Registry)
- GitHub workflow, monorepo management, and documentation best practices
apps/ # Frontend applications (React dashboard & HITL review UI)
services/ # FastAPI/agent microservices with independent Dockerfiles
libs/ # Reusable Python libraries (agent framework, RAG, auth, etc.)
supabase/ # Database schema, Edge Functions, migrations
scripts/ # Utility scripts for experiments and analytics
docs/ # Deep-dive design documents and architecture notes
- Start Docker Desktop:
open -a Docker - Launch core services:
make dev - Open the dashboard:
open http://localhost:8080
Services run with hot reload and connect to a remote Supabase instance — no local database required.
# Core development
make dev # start dashboard + backend + tools
make dev-logs # tail service logs
make dev-down # stop dev stack
make dev-rebuild # rebuild after dependency changes
# Full stack
make up # start ALL services + workers
make down # stop all containers
make logs s=<name> # inspect service logs
make ps # list running containers
# Testing
make test # run unit tests
make test-all # test every service
make smoke-test # end-to-end integration
make experiment-run # run experiment manifests
# Database
make migrate # local Alembic migrations
make migrate-prod # apply to Supabase
make db-shell # open psql shell
# Code quality
make fmt # format with ruff
make lint # lint with ruff
make lint-fix # auto-fix lint issues
# Deployment
make compose-cloud # local Cloud Run testing
make cloudrun-build # build Docker images
make cloudrun-push-all # push to Artifact RegistryThis repo is more than a toy application; it's a comprehensive platform that I designed, built, tested, and documented myself. It features:
- Architect complex systems with secure multi-tenant data handling.
- Integrate cutting-edge AI technologies and manage provider agnosticism.
- Build maintainable, reusable libraries supporting multiple services.
- Deliver production-ready code with observability and testing baked in.
- Collaborate using modern development workflows in a mono-repo structure.
Feel free to explore the code, run the stack locally, and reach out if you'd like a walkthrough or a chat about any part of the system.
Developed and maintained by Cid Lucas