<Sergio de la Garza/>

Full-stack BI platform for a logistics & fulfillment operator. Next.js 16 + React 19 frontend, Flask/Python REST API (19 endpoints, 18 optimized SQL views), BigQuery data layer across 5 datasets and 100+ source tables. Eight dashboards: Overview, Client Analytics, Freight & Carrier P&L, Order Analytics, Income Statement, Service Analytics, Warehouse & Storage, Year-over-Year. Sub-200ms API on large datasets.

• ▸ 2.26M transactions / $7.79M revenue tracked
• ▸ Python ETL on Cloud Run, 975K+ rows synced from Logiwa WMS
• ▸ Sub-200ms API on large datasets

Full-stack engineer on a funded startup. FastAPI/Python microservices, six named LangChain/LangGraph/CrewAI agents, AS2/EDI X12 protocol service with GCP KMS certificate management, Temporal.io durable workflows, ReactFlow visual pipeline builder, Monaco Editor in-browser code editing. Multi-model AI (OpenAI, Claude, Gemini, Vertex AI). Live integrations: QVC/Rithum, Tractor Supply, Cin7, ShipStation.

• ▸ 9 microservices, 6 autonomous AI agents
• ▸ AS2 + EDI X12 protocol service w/ GCP KMS
• ▸ Live integrations: QVC/Rithum, Tractor Supply, Cin7, ShipStation

Custom Model Context Protocol server (Node.js/TypeScript) enabling natural-language queries against Loganix's SPP business data — revenue, subscriptions, invoice analytics. Includes an ai-sql-generator tool so Claude can write and execute read-only SQL against the live MySQL database. Self-learning system that tracks successful query patterns. HTTP transport mode for remote access.

• ▸ Plain-English → SQL → live MySQL execution
• ▸ Non-technical staff query 100K+ records
• ▸ Self-learning pattern tracker

Python MCP server providing real-time AI-powered domain suggestions. Replaced a slow file-based system that generated 23MB+ JSON files with sub-second responses. OpenAI embeddings + Pinecone vector search + MySQL with 5 composite indexes for 10× query performance. Four MCP tools exposed to AI agents.

• ▸ 352K-domain vector index
• ▸ Replaced 23MB JSON file system → sub-second
• ▸ 10× query perf vs. previous solution

Claude Sonnet 4 generates SQL, executes it read-only, and returns results with automatic chart rendering (bar/line/area/pie via Recharts). AES-256-GCM encryption on stored credentials. SQL safety layer blocks all write operations. Built with Next.js 16, tRPC, Drizzle ORM, Vercel AI SDK. Docker Compose included with a 5K-customer demo database.

• ▸ Read-only SQL safety layer
• ▸ AES-256-GCM credential encryption
• ▸ Auto chart selection from result shape

Users connect accounts, import CSV transactions, and query financials in natural language — AI responses include automatic chart rendering via chartSpec JSONB. Next.js 16, tRPC 11, Drizzle ORM, Better Auth, Anthropic Claude. Docker Compose for self-hosting.

• ▸ Self-hostable via Docker Compose
• ▸ Chat → SQL → chart pipeline
• ▸ Open-source codebase

Built solo end-to-end. Multi-role user system, real-time automated tracking via headless browsers, live admin dashboard, Stripe billing.

• ▸ Real-time appointment tracking
• ▸ Multi-role auth + Stripe billing
• ▸ Live in production

Node.js + Puppeteer + puppeteer-extra-plugin-stealth (anti-detection), Socket.io real-time web UI, residential proxy rotation via Rayobyte, CAPTCHA handling, Electron desktop packaging, Mailgun email alerts.

• ▸ 10 consulates monitored concurrently
• ▸ Stealth Puppeteer + residential proxy rotation
• ▸ Electron desktop packaging