Meridian Meridian

Overview

Designed and scaffolded a low-code territory-intelligence SaaS that maps BigQuery-scale Mexican socio-demographic and geographic data into explainable geospatial applications.

Highlights

1. 01

   Designed a LangGraph orchestrator of 7 specialized agents (Profiler, Discovery, SQLGenerator and others) for text-to-SQL, with a critic loop that retries generated SQL up to twice.

2. 02

   Implemented a sqlglot-based SQL validator with six rules and four-tier BigQuery cost gates so every generated query is checked before it can run against a ~577M-row (~79 GB) data lake.

3. 03

   Set up a FastAPI gateway with RBAC, audit, PII-scan and quota services, Postgres + pgvector migrations, Terraform for Cloud Run and 5 architecture decision records.

Stack

• Next.js 14
• FastAPI
• LangGraph
• Google BigQuery
• Terraform
• Google Cloud Run
• GIS
• Geospatial Analysis
• Multi-Agent Systems
• sqlglot
• pgvector