Full-Stack & AI Engineer
Building production AI agents over live data. Currently shipping ChopBot — an OpenAI-function-calling agent grounded in a live Postgres of 100+ US grocery chains.
Core Strengths
AI Engineering
- ▹OpenAI Function Calling
- ▹RAG / Live-Data Grounding
- ▹Claude Code Workflows
- ▹MCP-aware Tooling
Full-Stack
- ▹Next.js 14
- ▹TypeScript
- ▹React
- ▹Python
- ▹REST APIs / SSE
Data & Caching
- ▹PostgreSQL
- ▹Redis (3-tier cache)
- ▹Supabase
- ▹Firestore
Cloud & DevOps
- ▹AWS (EC2, S3, CloudFront)
- ▹Vercel
- ▹Cloudflare
- ▹Git / CI/CD
AI Engineering Expertise
Production AI agents grounded in live data — function calling, RAG, multi-tool orchestration, and streaming UIs. Not just API calls; full infrastructure for AI-powered applications.
Use-Case Driven AI
- ▹Live-Data-Grounded Agent (RAG) — ChopBot grounds every response in a live Postgres of 100+ US grocery chains via OpenAI function calling, not training-set memory
- ▹Multi-Tool Orchestration — 8 custom tools (multi-chain search, price compare, history, store locator, etc.) chained dynamically with structured arguments
- ▹Streaming Conversational UI — Server-Sent Events for sub-1s first paint over the agent loop, tool calls resolving in parallel
- ▹Structured-Output Multimodal — Deterministic JSON outputs powering personalized PDF + image generation pipelines (Taroscoper)
Key Engineering Work
- ▹Function Calling at Scale — 8 deterministic tool schemas + a runtime SQL rewriter translating model arguments into safe parameterized Postgres queries
- ▹Three-Tier Caching — Redis layered for hot/warm/cold reads against the price catalog, sized for the agent loop's tool dispatch
- ▹Rate Limiting & Abuse Prevention — Per-IP throttle and bot detection on the agent endpoint; cost controls on the LLM layer
- ▹Daily Agentic Dev Cadence — Claude Code with CLAUDE.md context, custom slash commands, and structured planning → review → execute loops
AI Engineering Walkthroughs
Two production patterns I've shipped: an agentic loop with function calling and live-data RAG, and a structured-output multimodal pipeline.
ChopBot — Agentic Pattern
Function calling + RAG over live grocery data
Watch ChopBot orchestrate live tool calls against a Postgres database of 100+ US grocery chains, streaming the response in real time.
What you're seeing
- ▹Natural-language query parsed into structured tool calls via OpenAI function calling
- ▹Tools dispatched in parallel against live Postgres + a three-tier Redis cache
- ▹Tokens streamed via Server-Sent Events for sub-1s first paint
Taroscoper — Structured-Output Pattern
Question input → card draw → verdict + interpretation
Step 1: Question Input
Step 2: Card Draw
Step 3: Verdict + Interpretation
Architecture Snapshot
GroceryChop / ChopBotLead project — agentic loop with function calling and live-data RAG.
Featured Projects
View All →GroceryChop.com
AI-powered grocery price comparison platform. ChopBot is an OpenAI-function-calling agent with 8 custom tools wired to a live Postgres of 100+ US grocery chains across 50+ metros — a RAG-style architecture grounding every response in live data. Next.js 14 + TypeScript + Python scraping backend, three-tier Redis cache, SSE streaming. Currently shipping daily via Claude Code.
Taroscoper.com
AI-driven SaaS web platform for personalized tarot readings. Built with Next.js and Firebase, featuring authenticated users, Stripe payments, OpenAI API integration for AI chat and image generation, and personalized PDF reports. Scaled to ~3,000 monthly users, 1,000+ authenticated accounts, and 20k+ Instagram followers.
Runnit.us
Location-based platform connecting players through nearby discovery of public basketball courts, matches, and tournaments. Features include ELO rating system, tournament hosting, and a global court database. Built with web platform and React Native iOS prototype sharing Firebase data.