Timothy Michel
Full-Stack Software Engineer • Rails · Go · React · Docker • Agentic AI Systems with Anthropic SDK & Claude Haiku
West Covina, CA – Open to Remote (U.S.) | +1 626-998-2531 |
Professional Summary
Full-Stack Software Engineer with 3+ years of production experience designing and deploying secure, containerized applications — with a growing specialization in building production-grade agentic AI systems using the Anthropic SDK and Claude Haiku. Designs and implements autonomous tool-use loops where a large language model drives multi-step workflows against real external APIs, evaluates intermediate results, and iterates toward a goal without human intervention — going well beyond prompt engineering or AI-assisted code cleanup.
Specializes in Ruby on Rails and Go APIs, React/TypeScript SPAs, and Docker-based CI/CD pipelines from wireframe to production. Brings a disciplined engineering mindset to every layer of the stack: API design, SPA architecture, background processing, AI agent safety architecture, infrastructure automation, and deployment. Committed to building systems that are reliable, maintainable, and built to delight end users.
Agentic AI Development
Designs and ships production agentic AI systems using the Anthropic Ruby SDK and Claude Haiku 4.5 — where the model autonomously drives multi-step tool-use loops against live external APIs to satisfy complex, free-form user intent:
- Autonomous Tool-Use Loop (AgenticLocationSearch) — built a production venue-discovery agent where Claude Haiku receives an event organizer's free-form criteria ("cool vibes", "artsy with indoor seating", "good for a corporate mixer") and independently decides what to search for, how many searches to run, and when to stop. Claude is equipped with two tools —
search_venues(Google Places Text Search + Nearby Search fallback) andget_venue_details(Places Details API) — and drives an agentic loop for up to 8 turns via the official Anthropic Ruby SDK before emitting a curated, ranked result set - Tool Definition & Function Calling — authors rich JSON-schema tool specifications with natural-language descriptions that shape model behavior and search strategy; manages
tool_use/end_turnstop-reason routing across multi-turn message histories serialized as plain hashes for subsequent API turns - Criteria-Driven Ranking — Claude returns a structured JSON array where each venue carries a
criteria_matchedcount and a per-criterionreasonstring; server-side code sorts by match count then by Google rating and delivers up to 20 results to the React frontend as Google Maps push pins - Two-Layer Agent Safety Architecture — a pre-filter strips ineligible venue types and caches real Google Places type data by
place_idbefore Claude sees any results; a post-filter re-validates every venue in Claude's final output against that cache, ensuring the model cannot bypass type exclusions by omitting or altering fields in its JSON output - Three-Path Intelligent Routing — coordinates three distinct search strategies in a single service coordinator: named-venue keyword search (no AI), agentic Haiku search (criteria present), and category-based Nearby Search (no criteria); each path fails gracefully and falls back to the next
- Claude Code (Anthropic CLI) — additionally uses Claude Code throughout the development cycle for codebase analysis, targeted refactoring across Rails, Go, and React services, and systematic identification of technical debt in multi-service architectures
Core Technical Skills
Backend: Ruby 3 / Rails 7.1 (API-only architecture), Go (REST APIs), PostgreSQL 17 / PostGIS, Redis-backed Sidekiq (ActiveJob), Devise / Devise-JWT, RSpec
Frontend: React 18, TypeScript, Vite, Axios, SPA routing & state management, Google Maps JavaScript API, SCSS component architecture
Agentic AI: Anthropic Ruby SDK 1.36.0, Claude Haiku 4.5 (claude-haiku-4-5), agentic loop design, JSON-schema tool definitions, multi-turn history management, stop-reason routing (tool_use / end_turn), pre/post-filter safety nets, free-form criteria interpretation, Google Places API (Text Search, Nearby Search, Details)
DevOps & Infra: Docker / Docker Compose, GitHub Actions CI/CD, GitHub Container Registry (GHCR), Nginx reverse proxy, TLS termination, VPS deployment, SSH orchestration
AI Dev Tools: Claude Code (Anthropic CLI), prompt engineering for code generation, debugging, and architecture review. VS Code and NeoVim are editors of choice.
CMS & Platforms: WordPress, Joomla, GiveWP donation platform, Stripe payment integration, SMTP configuration
Media Systems: Jitsi conferencing platform, FFmpeg media processing pipelines.
Email processing: Comfortyable wth optimizing mail MX, SPS, DKIM, and DMARC records to ensure consistent delivery of emails.
Engineering Projects
Amigos Unite — Full-Stack Community Platform
Full-stack community event coordination platform built end-to-end from wireframe to production, integrating a security-hardened Rails API with advanced location intelligence, media processing, and resilient async infrastructure.
Backend Architecture
- API & Serialization: Rails 7 API-only application with object-oriented, convention-driven JSON output via
active_model_serializers,jbuilder, andjson-api-serializer; external API communication viaHTTParty - Authentication & Security: Stateless, scalable auth via Devise, Devise-JWT, and JWT; multi-provider login via OmniAuth; suspicious activity detection with Authtrail; brute-force login protection via Rack::Attack rate limiting; keyed-hash encryption of sensitive fields using Blind Index
- Location Intelligence & Media: Location search and geocoding via Geocoder and Google Maps Service Ruby; Google Cloud Vision for location-based image discovery; images fetched, formatted, and resized locally using Active Storage, Image Processing gem, and Ruby-Vips
- Infrastructure & Safety: Asynchronous background processing via Sidekiq and Redis (ActiveJob adapter); antivirus scanning of all uploaded files with Clamby
- Testing: Behavioral test coverage via RSpec to verify adherence to defined application behavior
- AI Tooling: Claude Code applied post-development to clean up and optimize the existing codebase
Frontend Architecture
- React + TypeScript SPA with AuthContext-based authentication flow and role-based UI gating
- Token-based SCSS architecture, Axios API integration, card/grid interface system
Infrastructure & Deployment
- Fully containerized Docker stack: API server, Sidekiq worker, PostgreSQL, Redis, and Nginx
- GitHub Actions CI/CD pipeline building Docker images and publishing to GHCR with automated VPS deployment via SSH orchestration
- Nginx reverse proxy with TLS termination; environment-isolated secrets management
Groovey Task Manager — Full-Stack Go + React Application
github.com/sagacic-tim/Todo_Axios_Api
Calendar-driven task management application demonstrating a clean, production-grade separation between API service, data persistence, frontend UI, and containerized infrastructure — modeled after real-world deployment pipelines.
- Backend (Go): RESTful API built with Gin (HTTP routing) and GORM (ORM) over PostgreSQL; strict layered architecture — transport (HTTP handlers) → service (business logic) → repository (data access) → models (persistence) — for testability and long-term maintainability
- Task Lifecycle: Full CRUD API (
GET,POST,PATCH,DELETE/api/tasks) managing tasks across four states: scheduled, rescheduled, completed, and dismissed; tasks include title, description, due date, and state - Frontend (React/TypeScript): Vite-built SPA presenting a calendar grid interface where tasks are organized by day; Axios API communication with popup editor interactions for inline task management
- Infrastructure: Fully containerized via Docker Compose (API, frontend, PostgreSQL); Nginx reverse proxy routing
/apito the Go service and/to the React SPA; Let’s Encrypt TLS; server-level Basic Auth protection for demo deployments - CI/CD: GitHub push → CI build → publish images to GHCR → VPS pulls and deploys via
docker compose up -d - Testing: Go test suite covering repository, service, and transport layers independently; SQLite in-memory databases used for fast, isolated test execution across all layers
Professional Experience
Full-Stack Developer
The Global Village Children’s Project — 2020 – Present
- Designed and implemented Stripe-based donation platform that generated over $40,000 in charitable contributions
- Migrated payment infrastructure from PayPal to Stripe under tight operational constraints, maintaining continuity of donation flows
- Built subscription lifecycle management and webhook processing pipeline via GiveWP integration
- Maintained WordPress infrastructure, deployment configuration, CI workflows, and SMTP email delivery systems
- Applied Claude Code to analyze and refactor legacy PHP/WordPress customizations, improving maintainability and reducing technical debt
Web Developer
Boston 9/11 Truth — 2018 – 2020
- Designed and maintained full website and content infrastructure for a media-focused organization
- Managed hosting configuration, platform operations, and deployment pipelines
Web Developer / Founding Member
9/11 Truth Action Project — 2014 – 2017
- Built website, donation platform, and member-facing web infrastructure from the ground up
- Integrated JWPlayer video streaming system delivering ad-free video content to subscribers
- Managed VPS hosting infrastructure, DNS configuration, and deployment operations
Education
Computer Science Coursework (Upper Division)
California State University, Fullerton
Physics Coursework (Upper Division)
California Polytechnic University, Pomona
Associate of Arts — Business Administration
Fullerton College