Products & Systems
Every project combines API-first development with integrations and automation so teams ship faster and operate with confidence.
Determination and rendering engine behind California statutory hazard disclosure covering the full range of disclosure products, built over a PostGIS and GeoServer spatial data infrastructure.
Stack: Python/FastAPI, PostGIS, GeoServer
Integrations: Celery, Prefect, WeasyPrint, FEMA NFHL, Terraform/AWS ECS
Order-to-delivery portal for hazard disclosure reports: address search with APN disambiguation, entitlement gating, checkout, and PDF delivery.
Stack: Next.js, TypeScript, Tailwind
Integrations: Zustand, JWT auth, portal-admin API, Atlas SDI report queue
Personal full-stack RAG platform: scripture-aware AI Q&A, social community, prayer journaling, devotionals, and a full admin console.
Stack: FastAPI, PostgreSQL/pgvector, Redis
Integrations: OpenAI GPT-4o, Google OAuth, TOTP, APScheduler
CRM-synced booking platform with location search, automated onboarding flows, and document packs triggered on placement milestones.
Stack: Next.js, Laravel, TypeScript
Integrations: Mapbox, Clerk, PandaDoc, Pipedrive
Onboarding is templated, bookings stay in sync with CRM records, and coordinators see placement status without manual reconciliation.
Booking-to-inspection pipeline integrating payments, job creation, and lifecycle comms in one reliable flow.
Stack: React, Node.js, TypeScript
Integrations: Stripe, ServiceM8, Brevo, Figma QA
Bookings reliably trigger payments, job templates, and lifecycle emails — cutting manual cleanup and eliminating sync failures.
Real-time resource management for responders with dispatch dashboards, movement audit trails, and RBAC.
Stack: Laravel, Inertia, Vue
Integrations: Inventory, dispatch workflows, reporting
Teams locate and dispatch resources faster with clear accountability — reducing coordination failures during incidents.
Unified HR platform linking attendance, payroll, and leave workflows with role-based access and audit exports.
Stack: Laravel, React, MySQL
Integrations: Payroll exports, role management, notifications
Attendance, leave, and payroll stay aligned with approvals and exports — removing manual reconciliation between HR and finance.
Conversion-focused storefront with Shopify backend, Checkout.com payments, and Meta Pixel on key funnel steps.
Stack: React, Shopify, Checkout.com
Integrations: Shopify, Checkout.com, Meta Pixel
Customers get a smooth checkout, orders sync cleanly to Shopify, and attribution is accurate — reducing cart drop-offs and support load.
Centralized inventory and shipment management with movement history, fulfillment dashboards, and staged rollouts.
Stack: ASP.NET, Angular, SQL Server
Integrations: Shipment tracking, product flow monitoring
Teams see inventory and shipment status in one place — reducing manual reconciliation and enabling faster fulfillment decisions.
6 internal production systems: document generation, anti-piracy fingerprinting, a court-record lead classification engine, multi-party dispute scheduling, a containerized scheduling service, and a Wear OS rule engine for mediator workflows.
Stack: React, Node.js/Express, TypeScript, Supabase
Integrations: n8n, SurveyJS, BullMQ, Docker, Wear OS
Featured breakdowns
Three builds in full — the problem each one solved, the decisions that shaped it, and what shipped. The rest are summarized above; happy to walk through any of them.
Context
Every residential property sale in California legally requires a Natural Hazard Disclosure. Producing one means checking a parcel against dozens of state and federal hazard datasets — flood, fire severity, seismic, landslide, liquefaction, dam inundation, airport noise — and emitting a document in a statutorily prescribed format. A slow report is an inconvenience; a wrong one is legal liability.
Your Role
Primary developer on the determination and reporting pipeline
Outcome
The full range of statutory disclosure products in production. Report generation moved from synchronous request-time rendering to a queued pipeline with warmup caching.
Frontend
Server-rendered PDF templates; report map and details pages consumed by the customer portal
Backend
FastAPI async Python over PostGIS/GeoAlchemy2, SQLAlchemy 2.0, Celery queue for async report orders
Infra
Terraform-defined AWS — ECS Fargate, RDS, ALB, Secrets Manager, ECR — plus Prefect flows promoting datasets from on-prem NAS to AWS
APIs & Integrations
GeoServer WMS/WFS, FEMA NFHL with local-copy failover, FAA ADHP, CAL FIRE FHSZ, USGS 3DEP, libpostal via gRPC
Key Decisions
Context
Escrow officers and agents ordering a hazard disclosure need to find one specific parcel — often from a partial address, sometimes with several APNs matching — then pick the right report type, and receive a PDF they can attach to a transaction. Getting the wrong parcel means the wrong disclosure on a legally binding sale.
Your Role
Sole developer — architecture, build, and QA
Outcome
Full order-to-delivery flow in production, and a marketing site rebuilt to Figma with the portal ported onto it.
Frontend
Next.js App Router with route groups, an AuthGuard app shell, Zustand stores, and a Tailwind component system built to Figma frames
Backend
Typed API client over the portal-admin JSON API and the Atlas SDI report endpoints
Infra
Multi-stage Docker with Next.js standalone output, non-root user, and a health endpoint; env-driven proxying to the SDI
APIs & Integrations
JWT auth with refresh, per-user report access limits, auto-generated escrow numbering, email delivery
Key Decisions
Context
Personal project to push production-grade RAG end-to-end. A Bible study app where users ask theological questions and receive scripture-grounded answers from a 14-stage pipeline that retrieves verses via pgvector similarity across 36,819 embeddings, enriches with Greek/Hebrew morphology, cross-references, and historical context, then personalizes responses using per-user memory and active prayer requests.
Your Role
Solo build — architecture, AI pipeline, backend, data modeling, DevOps, security
Outcome
Production-shaped backend (~3,400 LOC AI service, 20 routers, 25 services, 33 migrations, ~40 tables) demonstrating applied RAG, domain modeling depth, and operational maturity.
Frontend
Mobile/web client and admin dashboard as separate repos within the monorepo
Backend
FastAPI async Python, SQLAlchemy 2.0, Alembic (33 migrations), Pydantic v2
Infra
Docker Compose orchestrating 12 services with seed runners, profiles, and reset paths
APIs & Integrations
OpenAI GPT-4o / GPT-4o-mini / embeddings / moderation, Google OAuth, JWT + refresh-token rotation, pgvector, Redis, APScheduler, SlowAPI
Key Decisions
Full-Stack Developer · 6 Internal Systems · Active Production
No company name, links, or screenshots. Descriptions cover system architecture and outcomes only.
Legal forms are complex — static templates break the moment data changes. Built a form-driven pipeline where users complete a SurveyJS questionnaire embedded in the storefront, with answers auto-saving as they go. On submission the data triggers an n8n workflow that merges it into a Word template, converts to PDF, and delivers it by email. Paired with an internal admin app — built on the SurveyJS form builder — where staff manage forms, templates, and bundles without a developer, with all writes routed through a resource-style API rather than direct database access.
Zero manual document handling, and non-developers can ship new form types.
Self-help legal documents are trivially resold once they leave your system. Built a two-part fingerprinting subsystem: a Node/Express/TypeScript microservice that stamps formatted footers into .docx files over a REST API (bold, italic, color, font, alignment), and a companion service that injects per-purchase identity — user ID, email, timestamp — into the generated PDF's metadata via pdf-lib. Both are called from the document pipeline, so every delivered file is uniquely traceable to the buyer.
Every delivered document carries an invisible, per-buyer fingerprint.
Daily public court-record exports arrive as zipped, caret-delimited flat files with separate lookup tables for party types, case types, and judges. Built a Node/Express/TypeScript backend that parses the feed, filters to active family-law matters filed within 180 days, detects trigger events in the docket — service returned, motion to withdraw — and classifies each unrepresented party into one of four lead profiles with priority tiers: newly filed, recently served, dropped by counsel, or simply unrepresented. Leads are enriched with human-readable labels and batch-upserted behind a CRUD API.
A raw daily court feed becomes a prioritized, deduplicated lead queue with no manual triage.
Dispute negotiations require all three parties — creator, receiver, and mediator — to agree on a meeting time. Built a scheduling system where a creator proposes a date, the receiver accepts or declines, and on agreement all three parties are notified automatically. Designed for legal dispute workflows where neutral coordination matters.
Three-party consent model with automated notifications on agreement.
The scheduling workload outgrew a single process once reminders, confirmations, and calendar sync all needed to survive restarts. Split it into an Express API and a separate BullMQ worker over Redis, with Postgres for state and migrations applied automatically on container start. The whole stack — API, worker, database, cache, Traefik reverse proxy — ships as a documented compose deployment with a written production runbook.
Background jobs survive restarts; the full stack deploys from one documented runbook.
Mediators need to trigger timed actions during sessions — reminders, check-ins, delays — without interrupting the flow. Built a rule engine with three rule types: instant (fires immediately), delay (fires after a set timer), and cron (fires on a set schedule). Rules are configured in a frontend dashboard, mapped to named buttons, and surfaced on a Wear OS watch so the mediator can fire any rule with a single button press. Also includes a voice command system for natural-language scheduling.
Frontend rule configuration → one-tap execution from a watch during live sessions.